Erster Commit

2026-04-23 22:10:45 +02:00
commit 5d47482d2a
30 changed files with 6340 additions and 0 deletions
--- a/.claude/settings.local.json
+++ b/.claude/settings.local.json
@@ -0,0 +1,31 @@
+{
+  "permissions": {
+    "allow": [
+      "Bash(npm install *)",
+      "Bash(npx tsc *)",
+      "Bash(echo \"Exit: $?\")",
+      "Bash(npm run *)",
+      "Bash(pip install *)",
+      "Bash(pip show *)",
+      "Bash(pip3 show *)",
+      "Bash(pip3 list *)",
+      "Bash(.venv/bin/pip install *)",
+      "Bash(/opt/homebrew/opt/python@3.12/bin/python3.12 -m venv .venv --clear)",
+      "Bash(python3.13 -m venv .venv --clear)",
+      "Bash(.venv/bin/python -c \"import flask; import flask_cors; import ollama; import PIL; print\\('Alle Imports OK - Flask', flask.__version__\\)\")",
+      "Bash(.venv/bin/python app.py)",
+      "Bash(curl -s http://localhost:8000/api/images?mode=draw)",
+      "Bash(curl -s \"http://localhost:8000/api/images?mode=draw\")",
+      "Read(//tmp/**)",
+      "Bash(curl -s \"http://localhost:3000/\")",
+      "Bash(curl -s \"http://localhost:3000/api/images?mode=draw\")",
+      "mcp__coolify__ping",
+      "mcp__coolify__projects",
+      "mcp__coolify__servers",
+      "Bash(docker build *)",
+      "mcp__coolify__private-keys",
+      "Bash(pkill -f \"python app.py\")",
+      "Bash(curl -s http://localhost:8000/)"
+    ]
+  }
+}
--- a/.dockerignore
+++ b/.dockerignore
@@ -0,0 +1,23 @@
+# Python
+.venv/
+__pycache__/
+*.pyc
+*.pyo
+
+# Node
+frontend/node_modules/
+frontend/dist/
+
+# Nutzerdaten (werden als Volumes gemountet, nicht ins Image gepackt)
+pictures/
+objects_image/
+sentence_object/
+sentance_object/
+
+# Git & IDE
+.git/
+.gitignore
+.DS_Store
+
+# Bereits gebaute React-Dateien (werden im Docker-Build neu gebaut)
+static/react/
--- a/.gitignore
+++ b/.gitignore
@@ -0,0 +1,29 @@
+# Python
+.venv/
+__pycache__/
+*.pyc
+*.pyo
+*.pyd
+.env
+
+# Node
+frontend/node_modules/
+
+# Nutzerdaten (lokal bleiben, nicht ins Repo)
+pictures/
+objects_image/
+sentence_object/
+sentance_object/
+
+# Vite-Build (wird im Docker neu gebaut)
+static/react/
+
+# macOS
+.DS_Store
+
+# IDE
+.vscode/
+.idea/
+
+# Logs
+*.log
--- a/43
+++ b/43
@@ -0,0 +1,43 @@
+# ── Stage 1: React-Frontend bauen ───────────────────────────────────────────
+FROM node:20-alpine AS frontend-builder
+# /repo entspricht dem Repo-Root im Build-Context
+WORKDIR /repo/frontend
+COPY frontend/package*.json ./
+RUN npm ci --silent
+COPY frontend/ .
+# vite.config outDir: ../static/react → schreibt nach /repo/static/react
+RUN npm run build
+
+# ── Stage 2: Python/Flask Backend ────────────────────────────────────────────
+FROM python:3.12-slim
+WORKDIR /app
+
+# Bildcodec-Abhängigkeiten für Pillow
+RUN apt-get update && apt-get install -y --no-install-recommends \
+        libjpeg-dev libpng-dev libwebp-dev \
+    && rm -rf /var/lib/apt/lists/*
+
+# Python-Pakete
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+
+# Anwendungscode
+COPY app.py .
+COPY prompts/ prompts/
+COPY templates/ templates/
+
+# Leeres static/ vorbereiten, dann React-Build einfügen
+RUN mkdir -p static/react
+COPY --from=frontend-builder /repo/static/react ./static/react
+
+# Datenverzeichnisse (in Coolify als Volumes konfigurieren)
+RUN mkdir -p pictures objects_image sentence_object
+
+# Ollama-Verbindung: Standard → Ollama auf dem Host-Gerät
+# In Coolify als Umgebungsvariable überschreibbar
+ENV OLLAMA_HOST=http://host.docker.internal:11434
+ENV PYTHONUNBUFFERED=1
+
+EXPOSE 8000
+
+CMD ["gunicorn", "--bind", "0.0.0.0:8000", "--workers", "2", "--timeout", "120", "app:app"]
--- a/README.md
+++ b/README.md
@@ -0,0 +1,36 @@
+## Bild-Ausschnitt-Tool
+
+Kurzes Web-Tool, um ein lokales Bild aus dem Ordner `pictures` im Browser anzuzeigen, einen Rechteck-Ausschnitt mit der Maus zu wählen und den Ausschnitt in `objects_image` zu speichern.
+
+### Installation
+
+1. In den Projektordner wechseln:
+
+```bash
+cd /Users/tim/SynologyDrive/LanguageParent/content_mentor
+```
+
+2. (Optional, empfohlen) Virtuelle Umgebung anlegen/aktivieren und Abhängigkeiten installieren:
+
+```bash
+python -m venv .venv
+source .venv/bin/activate  # macOS / Linux
+pip install -r requirements.txt
+```
+
+3. Lege deine Quellbilder in den Ordner `pictures` neben dieser Datei.
+
+### Starten
+
+```bash
+python app.py
+```
+
+Dann im Browser aufrufen:
+
+```text
+http://localhost:5000
+```
+
+Den Port kannst du in `app.py` im `app.run(...)`-Aufruf anpassen und später per Nginx/Reverse-Proxy von außen erreichbar machen.
+
--- a/app.py
+++ b/app.py
@@ -0,0 +1,815 @@
+from pathlib import Path
+from datetime import datetime
+from uuid import uuid4
+import json
+
+from flask import Flask, send_from_directory, request, jsonify
+from flask_cors import CORS
+from PIL import Image
+import ollama
+
+BASE_DIR = Path(__file__).resolve().parent
+PICTURES_DIR = BASE_DIR / "pictures"
+OBJECTS_DIR = BASE_DIR / "objects_image"
+SENTENCE_DIR = BASE_DIR / "sentence_object"
+PROMPTS_DIR = BASE_DIR / "prompts"
+
+app = Flask(
+    __name__,
+    template_folder=str(BASE_DIR / "templates"),
+    static_folder=str(BASE_DIR / "static"),
+)
+CORS(app)
+
+
+def read_prompt(filepath: Path, fallback: str) -> str:
+    """
+    Hilfsfunktion, um Prompt-Dateien zu lesen.
+    """
+    try:
+        if filepath.exists():
+            return filepath.read_text(encoding="utf-8").strip()
+    except Exception:
+        pass
+    return fallback.strip()
+
+
+@app.route("/api/images", methods=["GET"])
+def list_images():
+    """
+    Gibt alle Bilder im pictures-Ordner zurück.
+    Query-Parameter: ?mode=draw (Standard) oder ?mode=generate
+    draw     → Bilder ohne _saved-Suffix
+    generate → Bilder mit _saved-Suffix
+    """
+    mode = request.args.get("mode", "draw")
+    PICTURES_DIR.mkdir(parents=True, exist_ok=True)
+    if mode == "generate":
+        image_paths = sorted(
+            [f for f in PICTURES_DIR.iterdir() if f.is_file() and f.stem.endswith("_saved")],
+            key=lambda p: p.stat().st_mtime,
+        )
+    else:
+        image_paths = sorted(
+            [f for f in PICTURES_DIR.iterdir() if f.is_file() and not f.stem.endswith("_saved")],
+            key=lambda p: p.stat().st_mtime,
+        )
+    return jsonify({"images": [p.name for p in image_paths]})
+
+
+REACT_BUILD_DIR = BASE_DIR / "static" / "react"
+
+
+def _serve_spa():
+    """Liefert die React SPA index.html für alle Frontend-Routen."""
+    return send_from_directory(REACT_BUILD_DIR, "index.html")
+
+
+@app.route("/")
+@app.route("/draw")
+@app.route("/generate")
+def serve_spa():
+    return _serve_spa()
+
+
+@app.route("/assets/<path:filename>")
+def serve_react_assets(filename: str):
+    """Liefert JS/CSS-Assets aus dem Vite-Build."""
+    return send_from_directory(REACT_BUILD_DIR / "assets", filename)
+
+
+@app.route("/pictures/<path:filename>")
+def serve_picture(filename: str):
+    # Statisches Ausliefern der Originalbilder
+    return send_from_directory(PICTURES_DIR, filename)
+
+
+@app.route("/objects_image/<path:filename>")
+def serve_object_image(filename: str):
+    # Ausliefern der gespeicherten Ausschnitt-Bilder
+    return send_from_directory(OBJECTS_DIR, filename)
+
+
+@app.route("/api/objects", methods=["GET"])
+def list_objects_for_image():
+    """
+    Gibt alle gespeicherten Objekte (Ausschnitte) zu einem Quellbild zurück.
+
+    Query-Parameter:
+      ?filename=testbild.png
+    """
+    filename = request.args.get("filename")
+    if not filename:
+        return jsonify({"error": "Missing filename query parameter"}), 400
+
+    # Für *_saved-Bilder auch Objekte berücksichtigen, deren source_filename
+    # noch auf den ursprünglichen Namen ohne "_saved" zeigt (Kompatibilität)
+    base_filename = filename
+    legacy_filename = None
+    if base_filename.endswith(".png") or base_filename.endswith(".jpg") or base_filename.endswith(".jpeg") or base_filename.endswith(".webp"):
+        stem = base_filename.rsplit(".", 1)[0]
+        suffix = base_filename.rsplit(".", 1)[1]
+        if stem.endswith("_saved"):
+            legacy_filename = f"{stem[:-6]}.{suffix}"
+
+    OBJECTS_DIR.mkdir(parents=True, exist_ok=True)
+    objects = []
+    for meta_file in OBJECTS_DIR.glob("*.txt"):
+        try:
+            meta = json.loads(meta_file.read_text(encoding="utf-8"))
+        except Exception:
+            continue
+
+        src_name = meta.get("source_filename")
+        if src_name != filename and (legacy_filename is None or src_name != legacy_filename):
+            continue
+
+        objects.append(
+            {
+                "id": meta.get("id"),
+                "image_file": meta.get("image_file"),
+                "title_de": meta.get("title_de", ""),
+                "position_de": meta.get("position_de", ""),
+                "action_de": meta.get("action_de", ""),
+                "condition_de": meta.get("condition_de", ""),
+                # Von KI erzeugte Details (optional)
+                "label_en": meta.get("label_en"),
+                "label_de": meta.get("label_de"),
+                "label_se": meta.get("label_se"),
+                "color_en": meta.get("color_en"),
+                "adjective_en": meta.get("adjective_en"),
+                "action_verb_en": meta.get("action_verb_en"),
+                "preposition_en": meta.get("preposition_en"),
+                "relative_position_en": meta.get("relative_position_en"),
+                "season_en": meta.get("season_en"),
+                "mode": meta.get("mode"),
+                "bbox": meta.get("bbox"),
+                "polygon": meta.get("polygon"),
+                "hierarchy": meta.get("hierarchy", 1),
+                "parent_id": meta.get("parent_id"),
+                "created_at": meta.get("created_at"),
+            }
+        )
+
+    # Älteste zuerst sortieren (1 = ältestes Objekt)
+    objects.sort(key=lambda o: o.get("created_at") or "")
+
+    # Laufende Nummer je Bild (1..n) nach obiger Sortierung
+    for idx, obj in enumerate(objects, start=1):
+        obj["index"] = idx
+
+    return jsonify({"objects": objects})
+
+
+@app.route("/api/object/<obj_id>/hierarchy", methods=["POST"])
+def update_object_hierarchy(obj_id: str):
+    """
+    Aktualisiert die Hierarchie (1, 2 oder 3) eines gespeicherten Objekts.
+    """
+    data = request.get_json(force=True, silent=True) or {}
+    try:
+        hierarchy = int(data.get("hierarchy"))
+    except (TypeError, ValueError):
+        return jsonify({"error": "Invalid hierarchy"}), 400
+
+    if hierarchy not in (1, 2, 3):
+        return jsonify({"error": "Hierarchy must be 1, 2 or 3"}), 400
+
+    meta_path = OBJECTS_DIR / f"{obj_id}.txt"
+    if not meta_path.exists():
+        return jsonify({"error": "Object not found"}), 404
+
+    try:
+        meta = json.loads(meta_path.read_text(encoding="utf-8"))
+    except Exception:
+        return jsonify({"error": "Could not read meta file"}), 500
+
+    meta["hierarchy"] = hierarchy
+    meta_path.write_text(json.dumps(meta, ensure_ascii=False, indent=2), encoding="utf-8")
+
+    return jsonify({"id": obj_id, "hierarchy": hierarchy})
+
+
+@app.route("/api/object/<obj_id>", methods=["POST"])
+def update_object_meta(obj_id: str):
+    """
+    Aktualisiert Text-Metadaten (title_de, position_de, action_de, condition_de)
+    eines gespeicherten Objekts.
+    """
+    data = request.get_json(force=True, silent=True) or {}
+
+    meta_path = OBJECTS_DIR / f"{obj_id}.txt"
+    if not meta_path.exists():
+        return jsonify({"error": "Object not found"}), 404
+
+    try:
+        meta = json.loads(meta_path.read_text(encoding="utf-8"))
+    except Exception:
+        return jsonify({"error": "Could not read meta file"}), 500
+
+    for key in ("title_de", "position_de", "action_de", "condition_de"):
+        if key in data:
+            meta[key] = data.get(key, "")
+
+    meta_path.write_text(json.dumps(meta, ensure_ascii=False, indent=2), encoding="utf-8")
+
+    return jsonify({k: meta.get(k) for k in ("id", "title_de", "position_de", "action_de", "condition_de")})
+
+
+@app.route("/api/object/<obj_id>/parent", methods=["POST"])
+def update_object_parent(obj_id: str):
+    """
+    Aktualisiert die Parent-Relation eines Objekts.
+    Erwartet JSON: { "parent_id": "<uuid>" | null }
+    """
+    data = request.get_json(force=True, silent=True) or {}
+    parent_id = data.get("parent_id")
+
+    meta_path = OBJECTS_DIR / f"{obj_id}.txt"
+    if not meta_path.exists():
+        return jsonify({"error": "Object not found"}), 404
+
+    try:
+        meta = json.loads(meta_path.read_text(encoding="utf-8"))
+    except Exception:
+        return jsonify({"error": "Could not read meta file"}), 500
+
+    # Optional: prüfen, ob parent_id existiert und zum selben Bild gehört
+    if parent_id:
+        parent_meta_path = OBJECTS_DIR / f"{parent_id}.txt"
+        if not parent_meta_path.exists():
+            return jsonify({"error": "Parent object not found"}), 400
+        try:
+            parent_meta = json.loads(parent_meta_path.read_text(encoding="utf-8"))
+        except Exception:
+            return jsonify({"error": "Could not read parent meta file"}), 500
+
+        if parent_meta.get("source_filename") != meta.get("source_filename"):
+            return jsonify({"error": "Parent must belong to same source image"}), 400
+
+    meta["parent_id"] = parent_id or None
+    meta_path.write_text(json.dumps(meta, ensure_ascii=False, indent=2), encoding="utf-8")
+
+    return jsonify({"id": obj_id, "parent_id": meta.get("parent_id")})
+
+
+@app.route("/api/object/<obj_id>/generate_details", methods=["POST"])
+def generate_object_details(obj_id: str):
+    """
+    Erzeugt mit Llama 3.2 Vision zusätzliche englische Metadaten zu einem Objekt
+    auf Basis des ausgeschnittenen Objektbildes + vorhandener deutscher Felder.
+    """
+    meta_path = OBJECTS_DIR / f"{obj_id}.txt"
+    if not meta_path.exists():
+        return jsonify({"error": "Object not found"}), 404
+
+    try:
+        meta = json.loads(meta_path.read_text(encoding="utf-8"))
+    except Exception:
+        return jsonify({"error": "Could not read meta file"}), 500
+
+    image_file = meta.get("image_file")
+    if not image_file:
+        return jsonify({"error": "Object has no image_file"}), 400
+
+    image_path = OBJECTS_DIR / image_file
+    if not image_path.exists():
+        return jsonify({"error": "Object image not found"}), 404
+
+    # Prompt laden
+    prompt_template = read_prompt(
+        PROMPTS_DIR / "create_details.txt",
+        "Analyze this object and return one JSON object with the requested fields.",
+    )
+
+    title_de = meta.get("title_de", "") or ""
+    action_de = meta.get("action_de", "") or ""
+    condition_de = meta.get("condition_de", "") or ""
+
+    # Platzhalter im Prompt ersetzen (einfache .format-Verwendung)
+    try:
+        prompt = prompt_template.format(
+            title_de=title_de,
+            action_de=action_de,
+            condition_de=condition_de,
+        )
+    except Exception:
+        # Falls das Format fehlschlägt, einfach Rohprompt + Kontext anhängen
+        prompt = (
+            f"{prompt_template}\n\n"
+            f"Titel (Deutsch): {title_de}\n"
+            f"Status (Deutsch): {action_de}\n"
+            f"Zustand (Deutsch): {condition_de}\n"
+        )
+
+    # Bild als Bytes laden
+    try:
+        img_bytes = image_path.read_bytes()
+    except Exception:
+        return jsonify({"error": "Could not read object image"}), 500
+
+    try:
+        response = ollama.chat(
+            model="llama3.2-vision",
+            messages=[{"role": "user", "content": prompt, "images": [img_bytes]}],
+            format="json",
+            options={"temperature": 0},
+        )
+    except Exception as e:
+        return jsonify({"error": f"LLM request failed: {e}"}), 500
+
+    content = response.get("message", {}).get("content")
+    if not content:
+        return jsonify({"error": "Empty response from LLM"}), 500
+
+    try:
+        ai_data = json.loads(content)
+    except Exception:
+        # Fallback: versuchen, das erste JSON-Objekt zu extrahieren
+        try:
+            start = content.index("{")
+            end = content.rindex("}") + 1
+            ai_data = json.loads(content[start:end])
+        except Exception:
+            return jsonify({"error": "Could not parse JSON from LLM response"}), 500
+
+    # Debug-Ausgabe des rohen AI-Outputs
+    try:
+        print(f"[generate_details] Raw AI data for {obj_id}: {json.dumps(ai_data, ensure_ascii=False)}")
+    except Exception:
+        print(f"[generate_details] Raw AI data for {obj_id} (non-serializable): {ai_data!r}")
+
+    # Erwartete Felder übernehmen (gemäß neuem Prompt)
+    fields = [
+        "label_en",
+        "label_de",
+        "label_se",
+        "color_en",
+        "adjective_en",
+        "action_verb_en",
+        "preposition_en",
+        "relative_position_en",
+        "season_en",
+    ]
+    for key in fields:
+        if key in ai_data:
+            meta[key] = ai_data.get(key)
+
+    # Metadatei zurückschreiben
+    meta_path.write_text(json.dumps(meta, ensure_ascii=False, indent=2), encoding="utf-8")
+
+    # Nur die neuen Felder zurückgeben (None als leeren String konvertieren)
+    result = {key: (meta.get(key) or "") for key in fields}
+
+    try:
+        print(f"[generate_details] Stored meta for {obj_id}: {json.dumps(result, ensure_ascii=False)}")
+    except Exception:
+        print(f"[generate_details] Stored meta for {obj_id}: {result!r}")
+
+    return jsonify(result)
+
+
+def load_sentences_for_object(obj_id: str):
+    """
+    Lädt alle gespeicherten Sätze für ein Objekt aus sentence_object/<obj_id>.txt.
+    Rückgabe: Liste von Dicts mit mindestens question_en, answer_en, created_at.
+    """
+    SENTENCE_DIR.mkdir(parents=True, exist_ok=True)
+    sentence_path = SENTENCE_DIR / f"{obj_id}.txt"
+    if not sentence_path.exists():
+        return []
+    try:
+        data = json.loads(sentence_path.read_text(encoding="utf-8"))
+        if isinstance(data, list):
+            return data
+        if isinstance(data, dict):
+            # Fallback: einzelnes Objekt in Liste wrappen
+            return [data]
+        return []
+    except Exception:
+        return []
+
+
+@app.route("/api/object/<obj_id>/sentences", methods=["GET"])
+def get_object_sentences(obj_id: str):
+    """
+    Gibt alle bisher erzeugten KI-Sätze zu einem Objekt zurück.
+    """
+    sentences = load_sentences_for_object(obj_id)
+    return jsonify({"object_id": obj_id, "sentences": sentences})
+
+
+@app.route("/api/object/<obj_id>/generate_sentence", methods=["POST"])
+def generate_object_sentence(obj_id: str):
+    """
+    Erzeugt mit Llama 3.1 einen neuen englischen Frage-Antwort-Satz
+    zu einem Objekt basierend auf den KI-Details und bisherigen Sätzen.
+    Speichert alle Sätze in sentence_object/<obj_id>.txt.
+    """
+    # Objekt-Metadaten laden (inkl. KI-Details)
+    meta_path = OBJECTS_DIR / f"{obj_id}.txt"
+    if not meta_path.exists():
+        return jsonify({"error": "Object not found"}), 404
+
+    try:
+        meta = json.loads(meta_path.read_text(encoding="utf-8"))
+    except Exception:
+        return jsonify({"error": "Could not read meta file"}), 500
+
+    # Relevante Felder für den Satz (KI-Details)
+    details = {
+        "label_en": meta.get("label_en"),
+        "label_de": meta.get("label_de"),
+        "label_se": meta.get("label_se"),
+        "color_en": meta.get("color_en"),
+        "adjective_en": meta.get("adjective_en"),
+        "action_verb_en": meta.get("action_verb_en"),
+        "preposition_en": meta.get("preposition_en"),
+        "relative_position_en": meta.get("relative_position_en"),
+        "season_en": meta.get("season_en"),
+        "title_de": meta.get("title_de"),
+        "position_de": meta.get("position_de"),
+        "action_de": meta.get("action_de"),
+        "condition_de": meta.get("condition_de"),
+    }
+
+    # Bisherige Sätze laden
+    previous_sentences = load_sentences_for_object(obj_id)
+
+    # Prompt laden
+    prompt_template = read_prompt(
+        PROMPTS_DIR / "create_sentence.txt",
+        "Create one new English question_en and answer_en as JSON, based on the object details and avoiding previous sentences.",
+    )
+
+    # Vollständigen Prompt zusammensetzen
+    try:
+        details_json = json.dumps(details, ensure_ascii=False, indent=2)
+        prev_json = json.dumps(previous_sentences, ensure_ascii=False, indent=2)
+    except Exception:
+        details_json = str(details)
+        prev_json = str(previous_sentences)
+
+    prompt = (
+        f"{prompt_template}\n\n"
+        f"OBJECT_DETAILS_JSON:\n{details_json}\n\n"
+        f"PREVIOUS_SENTENCES_JSON:\n{prev_json}\n"
+    )
+
+    # LLM-Aufruf (nur Text, kein Bild nötig)
+    try:
+        response = ollama.chat(
+            model="llama3.1",
+            messages=[{"role": "user", "content": prompt}],
+            format="json",
+            options={"temperature": 0.2},
+        )
+    except Exception as e:
+        return jsonify({"error": f"LLM request failed: {e}"}), 500
+
+    content = response.get("message", {}).get("content")
+    if not content:
+        return jsonify({"error": "Empty response from LLM"}), 500
+
+    try:
+        ai_data = json.loads(content)
+    except Exception:
+        # Fallback: erstes JSON-Objekt extrahieren
+        try:
+            start = content.index("{")
+            end = content.rindex("}") + 1
+            ai_data = json.loads(content[start:end])
+        except Exception:
+            return jsonify({"error": "Could not parse JSON from LLM response"}), 500
+
+    question_simple = (ai_data.get("question_simple_en") or "").strip()
+    answer_simple = (ai_data.get("answer_simple_en") or "").strip()
+    question_advanced = (ai_data.get("question_advanced_en") or "").strip()
+    answer_advanced = (ai_data.get("answer_advanced_en") or "").strip()
+    
+    if not question_simple or not answer_simple or not question_advanced or not answer_advanced:
+        return jsonify({"error": "LLM response missing required fields (question_simple_en, answer_simple_en, question_advanced_en, answer_advanced_en)"}), 500
+
+    # Neuen Satz mit Timestamp anreichern
+    entry = {
+        "object_id": obj_id,
+        "created_at": datetime.now().isoformat(),
+        "question_simple_en": question_simple,
+        "answer_simple_en": answer_simple,
+        "question_advanced_en": question_advanced,
+        "answer_advanced_en": answer_advanced,
+    }
+
+    sentences = previous_sentences + [entry]
+
+    # In Datei speichern
+    SENTENCE_DIR.mkdir(parents=True, exist_ok=True)
+    sentence_path = SENTENCE_DIR / f"{obj_id}.txt"
+    sentence_path.write_text(json.dumps(sentences, ensure_ascii=False, indent=2), encoding="utf-8")
+
+    return jsonify({"object_id": obj_id, "sentence": entry, "count": len(sentences)})
+
+
+@app.route("/api/image/save", methods=["POST"])
+def save_image():
+    """
+    Markiert ein Bild und alle zugehörigen Objekte als 'gespeichert',
+    indem der Dateiname um '_saved' erweitert wird und die Metadaten
+    der Objekte angepasst werden.
+    """
+    data = request.get_json(force=True, silent=True) or {}
+    filename = data.get("filename")
+    if not filename:
+        return jsonify({"error": "Missing filename"}), 400
+
+    src_path = PICTURES_DIR / filename
+    if not src_path.exists():
+        return jsonify({"error": "Image not found"}), 404
+
+    if src_path.stem.endswith("_saved"):
+        return jsonify({"error": "Image already saved"}), 400
+
+    new_name = f"{src_path.stem}_saved{src_path.suffix}"
+    dst_path = PICTURES_DIR / new_name
+
+    # Bild umbenennen
+    src_path.rename(dst_path)
+
+    # Zugehörige Objekte aktualisieren
+    OBJECTS_DIR.mkdir(parents=True, exist_ok=True)
+    for meta_file in OBJECTS_DIR.glob("*.txt"):
+        try:
+            meta = json.loads(meta_file.read_text(encoding="utf-8"))
+        except Exception:
+            continue
+
+        if meta.get("source_filename") != filename:
+            continue
+
+        meta["source_filename"] = new_name
+        meta_file.write_text(
+            json.dumps(meta, ensure_ascii=False, indent=2),
+            encoding="utf-8",
+        )
+
+    return jsonify({"old_name": filename, "new_name": new_name})
+
+
+@app.route("/api/crop", methods=["POST"])
+def crop_image():
+    """
+    Erwartet JSON:
+    {
+      "filename": "bild.jpg",
+      "selections": [
+        {
+          "number": 1,
+          "mode": "rect" | "polygon",
+          "bbox": {"x": 10, "y": 20, "width": 200, "height": 150},  // für rect
+          "polygon": [{"x": 10, "y": 20}, ...]  // für polygon
+        },
+        ...
+      ],
+      "title_de": "...",
+      "position_de": "...",
+      "action_de": "...",
+      "condition_de": "..."
+    }
+    
+    ODER (Legacy-Format für Kompatibilität):
+    {
+      "filename": "bild.jpg",
+      "mode": "rect" | "polygon",
+      "x": 10, "y": 20, "width": 200, "height": 150,  // für rect
+      "polygon": [{"x": 10, "y": 20}, ...]  // für polygon
+    }
+    """
+    data = request.get_json(force=True, silent=True) or {}
+
+    if "filename" not in data:
+        return jsonify({"error": "Missing filename"}), 400
+
+    src_path = PICTURES_DIR / data["filename"]
+    if src_path.stem.endswith("_saved"):
+        return jsonify({"error": "Cannot crop on saved image"}), 400
+    if not src_path.exists():
+        return jsonify({"error": "Source image not found"}), 404
+
+    OBJECTS_DIR.mkdir(parents=True, exist_ok=True)
+
+    # UUID für Bild + Metadaten
+    obj_id = uuid4().hex
+    timestamp = datetime.now().isoformat()
+    out_image_name = f"{obj_id}{src_path.suffix}"
+    out_image_path = OBJECTS_DIR / out_image_name
+    out_meta_path = OBJECTS_DIR / f"{obj_id}.txt"
+
+    # Prüfen, ob neues Format (selections) oder Legacy-Format
+    selections = data.get("selections")
+    
+    if selections and isinstance(selections, list) and len(selections) > 0:
+        # Neues Format: mehrere Auswahlen
+        processed_selections = []
+        
+        with Image.open(src_path) as img:
+            img_w, img_h = img.size
+            
+            for sel in selections:
+                sel_mode = sel.get("mode")
+                if sel_mode == "rect":
+                    bbox = sel.get("bbox")
+                    if not bbox:
+                        continue
+                    try:
+                        x = int(bbox.get("x", 0))
+                        y = int(bbox.get("y", 0))
+                        w = int(bbox.get("width", 0))
+                        h = int(bbox.get("height", 0))
+                    except (ValueError, TypeError):
+                        continue
+                    
+                    if w <= 0 or h <= 0:
+                        continue
+                    
+                    x2 = min(x + w, img_w)
+                    y2 = min(y + h, img_h)
+                    x1 = max(0, x)
+                    y1 = max(0, y)
+                    
+                    if x1 >= x2 or y1 >= y2:
+                        continue
+                    
+                    processed_selections.append({
+                        "number": sel.get("number", len(processed_selections) + 1),
+                        "mode": "rect",
+                        "bbox": {"x": x1, "y": y1, "width": x2 - x1, "height": y2 - y1},
+                    })
+                    
+                elif sel_mode == "polygon":
+                    polygon = sel.get("polygon")
+                    if not isinstance(polygon, list) or len(polygon) < 3:
+                        continue
+                    
+                    try:
+                        xs = [int(p.get("x", 0)) for p in polygon]
+                        ys = [int(p.get("y", 0)) for p in polygon]
+                    except (KeyError, TypeError, ValueError):
+                        continue
+                    
+                    min_x = max(min(xs), 0)
+                    min_y = max(min(ys), 0)
+                    max_x = min(max(xs), img_w)
+                    max_y = min(max(ys), img_h)
+                    
+                    if min_x >= max_x or min_y >= max_y:
+                        continue
+                    
+                    processed_selections.append({
+                        "number": sel.get("number", len(processed_selections) + 1),
+                        "mode": "polygon",
+                        "polygon": polygon,
+                        "bbox": {"x": min_x, "y": min_y, "width": max_x - min_x, "height": max_y - min_y},
+                    })
+        
+        if not processed_selections:
+            return jsonify({"error": "No valid selections provided"}), 400
+        
+        # Erstes Bild aus erster Auswahl erstellen
+        first_sel = processed_selections[0]
+        try:
+            with Image.open(src_path) as img:
+                if first_sel["mode"] == "rect":
+                    bbox = first_sel["bbox"]
+                    x1 = bbox["x"]
+                    y1 = bbox["y"]
+                    x2 = x1 + bbox["width"]
+                    y2 = y1 + bbox["height"]
+                    cropped = img.crop((x1, y1, x2, y2))
+                else:  # polygon - verwende Bounding-Box für das Bild
+                    bbox = first_sel["bbox"]
+                    x1 = bbox["x"]
+                    y1 = bbox["y"]
+                    x2 = x1 + bbox["width"]
+                    y2 = y1 + bbox["height"]
+                    cropped = img.crop((x1, y1, x2, y2))
+                
+                # Bild speichern (Format beibehalten)
+                img_format = img.format or "PNG"
+                cropped.save(out_image_path, format=img_format)
+                print(f"[crop_image] Bild gespeichert: {out_image_path} (Größe: {cropped.size}, Format: {img_format})")
+        except Exception as e:
+            print(f"[crop_image] Fehler beim Erstellen des Bildes: {e}")
+            return jsonify({"error": f"Failed to create image: {e}"}), 500
+        
+        # Metadaten mit allen Auswahlen speichern
+        meta = {
+            "id": obj_id,
+            "created_at": timestamp,
+            "source_filename": data["filename"],
+            "mode": processed_selections[0]["mode"],  # Legacy-Feld
+            "bbox": processed_selections[0]["bbox"],  # Legacy-Feld
+            "image_file": out_image_name,
+            "hierarchy": 1,
+            "parent_id": None,
+            "title_de": data.get("title_de", ""),
+            "position_de": data.get("position_de", ""),
+            "action_de": data.get("action_de", ""),
+            "condition_de": data.get("condition_de", ""),
+            "selections": processed_selections,  # Neue Feld: alle Auswahlen nummeriert
+        }
+        if processed_selections[0]["mode"] == "polygon":
+            meta["polygon"] = processed_selections[0]["polygon"]  # Legacy-Feld
+        
+    else:
+        # Legacy-Format: einzelne Auswahl (für Rückwärtskompatibilität)
+        mode = data.get("mode", "rect")
+        if mode not in {"rect", "polygon"}:
+            return jsonify({"error": "Invalid mode"}), 400
+
+        with Image.open(src_path) as img:
+            img_w, img_h = img.size
+            if mode == "rect":
+                try:
+                    x = int(data["x"])
+                    y = int(data["y"])
+                    w = int(data["width"])
+                    h = int(data["height"])
+                except (ValueError, TypeError, KeyError):
+                    return jsonify({"error": "Invalid rectangle coordinates"}), 400
+
+                if w <= 0 or h <= 0:
+                    return jsonify({"error": "Width and height must be positive"}), 400
+
+                x2 = min(x + w, img_w)
+                y2 = min(y + h, img_h)
+                x1 = max(0, x)
+                y1 = max(0, y)
+
+                if x1 >= x2 or y1 >= y2:
+                    return jsonify({"error": "Crop area outside of image"}), 400
+
+                cropped = img.crop((x1, y1, x2, y2))
+                bbox = {"x": x1, "y": y1, "width": x2 - x1, "height": y2 - y1}
+            else:  # polygon
+                polygon = data.get("polygon") or []
+                if not isinstance(polygon, list) or len(polygon) < 3:
+                    return jsonify({"error": "Polygon must have at least 3 points"}), 400
+
+                try:
+                    xs = [int(p["x"]) for p in polygon]
+                    ys = [int(p["y"]) for p in polygon]
+                except (KeyError, TypeError, ValueError):
+                    return jsonify({"error": "Invalid polygon coordinates"}), 400
+
+                min_x = max(min(xs), 0)
+                min_y = max(min(ys), 0)
+                max_x = min(max(xs), img_w)
+                max_y = min(max(ys), img_h)
+
+                if min_x >= max_x or min_y >= max_y:
+                    return jsonify({"error": "Polygon bounding box outside of image"}), 400
+
+                cropped = img.crop((min_x, min_y, max_x, max_y))
+                bbox = {
+                    "x": min_x,
+                    "y": min_y,
+                    "width": max_x - min_x,
+                    "height": max_y - min_y,
+                }
+
+            cropped.save(out_image_path)
+
+        # Metadaten als JSON in .txt speichern
+        meta = {
+            "id": obj_id,
+            "created_at": timestamp,
+            "source_filename": data["filename"],
+            "mode": mode,
+            "bbox": bbox,
+            "image_file": out_image_name,
+            "hierarchy": 1,
+            "parent_id": None,
+            "title_de": data.get("title_de", ""),
+            "position_de": data.get("position_de", ""),
+            "action_de": data.get("action_de", ""),
+            "condition_de": data.get("condition_de", ""),
+        }
+        if mode == "polygon":
+            meta["polygon"] = data.get("polygon")
+
+    out_meta_path.write_text(json.dumps(meta, ensure_ascii=False, indent=2), encoding="utf-8")
+
+    return jsonify(
+        {
+            "id": obj_id,
+            "image_file": out_image_name,
+            "meta_file": out_meta_path.name,
+        }
+    )
+
+
+if __name__ == "__main__":
+    # Du kannst den Port hier anpassen, z.B. 8000 oder 5000.
+    # Für externen Zugriff später einfach in Nginx/Reverse Proxy weiterleiten.
+    app.run(host="0.0.0.0", port=8000, debug=True)
+
--- a/frontend/index.html
+++ b/frontend/index.html
@@ -0,0 +1,12 @@
+<!DOCTYPE html>
+<html lang="de">
+  <head>
+    <meta charset="UTF-8" />
+    <meta name="viewport" content="width=device-width, initial-scale=1.0" />
+    <title>Content Mentor</title>
+  </head>
+  <body>
+    <div id="root"></div>
+    <script type="module" src="/src/main.tsx"></script>
+  </body>
+</html>
--- a/frontend/package-lock.json
+++ b/frontend/package-lock.json
--- a/frontend/package.json
+++ b/frontend/package.json
@@ -0,0 +1,23 @@
+{
+  "name": "content-mentor-frontend",
+  "private": true,
+  "version": "0.1.0",
+  "type": "module",
+  "scripts": {
+    "dev": "vite",
+    "build": "tsc && vite build",
+    "preview": "vite preview"
+  },
+  "dependencies": {
+    "react": "^18.3.1",
+    "react-dom": "^18.3.1",
+    "react-router-dom": "^6.26.0"
+  },
+  "devDependencies": {
+    "@types/react": "^18.3.5",
+    "@types/react-dom": "^18.3.0",
+    "@vitejs/plugin-react": "^4.3.1",
+    "typescript": "^5.5.3",
+    "vite": "^5.4.1"
+  }
+}
--- a/frontend/src/App.tsx
+++ b/frontend/src/App.tsx
@@ -0,0 +1,13 @@
+import { Routes, Route, Navigate } from 'react-router-dom'
+import DrawIt from './pages/DrawIt'
+import GenerateIt from './pages/GenerateIt'
+
+export default function App() {
+  return (
+    <Routes>
+      <Route path="/" element={<Navigate to="/draw" replace />} />
+      <Route path="/draw" element={<DrawIt />} />
+      <Route path="/generate" element={<GenerateIt />} />
+    </Routes>
+  )
+}
--- a/frontend/src/api.ts
+++ b/frontend/src/api.ts
@@ -0,0 +1,106 @@
+import type { ObjectMeta, Sentence } from './types'
+
+export async function getImages(mode: 'draw' | 'generate'): Promise<string[]> {
+  const res = await fetch(`/api/images?mode=${mode}`)
+  if (!res.ok) throw new Error('Fehler beim Laden der Bilder')
+  const data = await res.json()
+  return data.images as string[]
+}
+
+export async function getObjects(filename: string): Promise<ObjectMeta[]> {
+  const res = await fetch(`/api/objects?filename=${encodeURIComponent(filename)}`)
+  if (!res.ok) throw new Error('Fehler beim Laden der Objekte')
+  const data = await res.json()
+  return (data.objects || []) as ObjectMeta[]
+}
+
+export async function cropImage(payload: {
+  filename: string
+  selections: Array<{
+    number: number
+    mode: string
+    bbox?: { x: number; y: number; width: number; height: number } | null
+    polygon?: Array<{ x: number; y: number }> | null
+  }>
+  title_de: string
+  position_de: string
+  action_de: string
+  condition_de: string
+}): Promise<{ id: string; image_file: string; meta_file: string }> {
+  const res = await fetch('/api/crop', {
+    method: 'POST',
+    headers: { 'Content-Type': 'application/json' },
+    body: JSON.stringify(payload),
+  })
+  const data = await res.json()
+  if (!res.ok) throw new Error(data.error || 'Fehler beim Speichern des Ausschnitts')
+  return data
+}
+
+export async function saveImage(filename: string): Promise<{ old_name: string; new_name: string }> {
+  const res = await fetch('/api/image/save', {
+    method: 'POST',
+    headers: { 'Content-Type': 'application/json' },
+    body: JSON.stringify({ filename }),
+  })
+  const data = await res.json()
+  if (!res.ok) throw new Error(data.error || 'Fehler beim Speichern des Bildes')
+  return data
+}
+
+export async function updateObjectMeta(
+  objId: string,
+  meta: { title_de: string; position_de: string; action_de: string; condition_de: string }
+): Promise<void> {
+  const res = await fetch(`/api/object/${encodeURIComponent(objId)}`, {
+    method: 'POST',
+    headers: { 'Content-Type': 'application/json' },
+    body: JSON.stringify(meta),
+  })
+  if (!res.ok) throw new Error('Fehler beim Aktualisieren der Metadaten')
+}
+
+export async function updateHierarchy(objId: string, hierarchy: number): Promise<void> {
+  const res = await fetch(`/api/object/${encodeURIComponent(objId)}/hierarchy`, {
+    method: 'POST',
+    headers: { 'Content-Type': 'application/json' },
+    body: JSON.stringify({ hierarchy }),
+  })
+  if (!res.ok) throw new Error('Fehler beim Aktualisieren der Hierarchie')
+}
+
+export async function updateParent(objId: string, parentId: string | null): Promise<void> {
+  const res = await fetch(`/api/object/${encodeURIComponent(objId)}/parent`, {
+    method: 'POST',
+    headers: { 'Content-Type': 'application/json' },
+    body: JSON.stringify({ parent_id: parentId }),
+  })
+  if (!res.ok) throw new Error('Fehler beim Aktualisieren der Parent-Relation')
+}
+
+export async function generateDetails(objId: string): Promise<Partial<ObjectMeta>> {
+  const res = await fetch(`/api/object/${encodeURIComponent(objId)}/generate_details`, {
+    method: 'POST',
+  })
+  const data = await res.json()
+  if (!res.ok) throw new Error(data.error || 'Fehler bei KI-Details')
+  return data
+}
+
+export async function getSentences(objId: string): Promise<Sentence[]> {
+  const res = await fetch(`/api/object/${encodeURIComponent(objId)}/sentences`)
+  if (!res.ok) throw new Error('Fehler beim Laden der Sätze')
+  const data = await res.json()
+  return (data.sentences || []) as Sentence[]
+}
+
+export async function generateSentence(
+  objId: string
+): Promise<{ sentence: Sentence; count: number }> {
+  const res = await fetch(`/api/object/${encodeURIComponent(objId)}/generate_sentence`, {
+    method: 'POST',
+  })
+  const data = await res.json()
+  if (!res.ok) throw new Error(data.error || 'Fehler bei KI-Sentence')
+  return data
+}
--- a/frontend/src/components/DetailsPanel.tsx
+++ b/frontend/src/components/DetailsPanel.tsx
@@ -0,0 +1,71 @@
+import type { ObjectMeta, Sentence } from '../types'
+
+interface Props {
+  obj: ObjectMeta | null
+  objects: ObjectMeta[]
+  sentences: Sentence[]
+}
+
+function DetailRow({ label, value }: { label: string; value?: string }) {
+  return (
+    <div className="sidebar-row">
+      <label>{label}</label>
+      <div className="detail-value">{value || ''}</div>
+    </div>
+  )
+}
+
+export default function DetailsPanel({ obj, objects, sentences }: Props) {
+  const latestSentence = sentences.length > 0 ? sentences[sentences.length - 1] : null
+
+  const parentDisplay = obj?.parent_id
+    ? (() => {
+        const parent = objects.find(o => o.id === obj.parent_id)
+        return parent ? `${parent.index} - ${parent.title_de || 'ohne Titel'}` : ''
+      })()
+    : ''
+
+  return (
+    <>
+      <div className="sidebar-section">
+        <h2>Details</h2>
+        <DetailRow label="Titel" value={obj?.title_de} />
+        <DetailRow label="Position" value={obj?.position_de} />
+        <DetailRow label="Status (sitzt/schwimmt/segelt)" value={obj?.action_de} />
+        <DetailRow label="Zustand (alt/jung/rostig)" value={obj?.condition_de} />
+        <DetailRow label="Hierarchie" value={obj?.hierarchy != null ? String(obj.hierarchy) : ''} />
+        <div className="sidebar-row">
+          <label>Gehört zu (Parent-Index)</label>
+          <div className="detail-value">{parentDisplay}</div>
+        </div>
+      </div>
+
+      <div className="sidebar-section">
+        <h2>KI-Details</h2>
+        <DetailRow label="Label (EN)" value={obj?.label_en} />
+        <DetailRow label="Label (DE)" value={obj?.label_de} />
+        <DetailRow label="Label (SE)" value={obj?.label_se} />
+        <DetailRow label="Farbe (EN)" value={obj?.color_en} />
+        <DetailRow label="Adjektiv (EN)" value={obj?.adjective_en} />
+        <DetailRow label="Action Verb (EN)" value={obj?.action_verb_en} />
+        <DetailRow label="Präposition (EN)" value={obj?.preposition_en} />
+        <DetailRow label="Relative Position (EN)" value={obj?.relative_position_en} />
+        <DetailRow label="Season (EN)" value={obj?.season_en} />
+      </div>
+
+      <div className="sidebar-section">
+        <h2>KI-Sentence</h2>
+        <h3 style={{ fontSize: '0.9rem', marginTop: 0, marginBottom: 8, color: '#475569' }}>
+          Einfach (für Kinder)
+        </h3>
+        <DetailRow label="Question (EN)" value={latestSentence?.question_simple_en} />
+        <DetailRow label="Answer (EN)" value={latestSentence?.answer_simple_en} />
+        <h3 style={{ fontSize: '0.9rem', marginTop: 12, marginBottom: 8, color: '#475569' }}>
+          Fortgeschritten
+        </h3>
+        <DetailRow label="Question (EN)" value={latestSentence?.question_advanced_en} />
+        <DetailRow label="Answer (EN)" value={latestSentence?.answer_advanced_en} />
+      </div>
+    </>
+  )
+}
--- a/frontend/src/components/DrawCanvas.tsx
+++ b/frontend/src/components/DrawCanvas.tsx
@@ -0,0 +1,338 @@
+import {
+  forwardRef,
+  useCallback,
+  useEffect,
+  useImperativeHandle,
+  useRef,
+} from 'react'
+import type { ObjectMeta, Point, Selection } from '../types'
+
+export interface DrawCanvasHandle {
+  getCurrentSelection: () => Selection | null
+  resetSelection: () => void
+}
+
+interface Props {
+  imageSrc: string | null
+  objects: ObjectMeta[]
+  selectedObjectId: string | null
+  mode: 'rect' | 'polygon'
+  onHasSelection: (has: boolean) => void
+}
+
+export default forwardRef<DrawCanvasHandle, Props>(function DrawCanvas(
+  { imageSrc, objects, selectedObjectId, mode, onHasSelection },
+  ref
+) {
+  const canvasRef = useRef<HTMLCanvasElement>(null)
+
+  // Drawing state in refs — no re-renders needed
+  const isDragging = useRef(false)
+  const startXY = useRef({ x: 0, y: 0 })
+  const currentXY = useRef({ x: 0, y: 0 })
+  const polygonPoints = useRef<Point[]>([])
+  const isPolygonClosed = useRef(false)
+  const displayScale = useRef(1)
+  const imageRef = useRef<HTMLImageElement | null>(null)
+
+  // Keep latest props accessible from stable callbacks
+  const modeRef = useRef(mode)
+  const objectsRef = useRef(objects)
+  const selectedObjectIdRef = useRef(selectedObjectId)
+  const onHasSelectionRef = useRef(onHasSelection)
+
+  useEffect(() => { modeRef.current = mode }, [mode])
+  useEffect(() => { objectsRef.current = objects }, [objects])
+  useEffect(() => { selectedObjectIdRef.current = selectedObjectId }, [selectedObjectId])
+  useEffect(() => { onHasSelectionRef.current = onHasSelection }, [onHasSelection])
+
+  const redraw = useCallback(() => {
+    const canvas = canvasRef.current
+    const ctx = canvas?.getContext('2d')
+    if (!canvas || !ctx) return
+
+    ctx.clearRect(0, 0, canvas.width, canvas.height)
+
+    const img = imageRef.current
+    if (img) ctx.drawImage(img, 0, 0, canvas.width, canvas.height)
+
+    const scale = displayScale.current
+    const selectedId = selectedObjectIdRef.current
+
+    // Draw saved objects
+    for (const obj of objectsRef.current) {
+      if (obj.visible === false) continue
+      const isSelected = obj.id === selectedId
+      const h = obj.hierarchy || 1
+
+      let stroke = '#14532d'
+      let fill = 'rgba(20, 83, 45, 0.2)'
+      if (h === 1) { stroke = '#6b7280'; fill = 'rgba(107, 114, 128, 0.2)' }
+      else if (h === 2) { stroke = '#eab308'; fill = 'rgba(234, 179, 8, 0.3)' }
+      else if (h === 3) { stroke = '#dc2626'; fill = 'rgba(220, 38, 38, 0.3)' }
+
+      ctx.save()
+      ctx.strokeStyle = stroke
+      ctx.fillStyle = fill
+      ctx.lineWidth = isSelected ? 3 : 2
+      ctx.setLineDash(isSelected ? [2, 2] : [4, 3])
+
+      const { polygon, bbox } = obj
+      if (polygon && polygon.length >= 3) {
+        ctx.beginPath()
+        ctx.moveTo(polygon[0].x * scale, polygon[0].y * scale)
+        for (let i = 1; i < polygon.length; i++) ctx.lineTo(polygon[i].x * scale, polygon[i].y * scale)
+        ctx.closePath()
+        ctx.fill()
+        ctx.stroke()
+      } else if (bbox) {
+        ctx.fillRect(bbox.x * scale, bbox.y * scale, bbox.width * scale, bbox.height * scale)
+        ctx.strokeRect(bbox.x * scale, bbox.y * scale, bbox.width * scale, bbox.height * scale)
+      }
+
+      // White highlight ring for selected object
+      if (isSelected) {
+        ctx.strokeStyle = '#ffffff'
+        ctx.lineWidth = 2
+        ctx.setLineDash([])
+        if (polygon && polygon.length >= 3) {
+          ctx.beginPath()
+          ctx.moveTo(polygon[0].x * scale, polygon[0].y * scale)
+          for (let i = 1; i < polygon.length; i++) ctx.lineTo(polygon[i].x * scale, polygon[i].y * scale)
+          ctx.closePath()
+          ctx.stroke()
+        } else if (bbox) {
+          ctx.strokeRect(bbox.x * scale, bbox.y * scale, bbox.width * scale, bbox.height * scale)
+        }
+      }
+
+      // Index number in object center
+      const indexLabel = typeof obj.index === 'number' ? String(obj.index) : ''
+      if (indexLabel) {
+        let cx = 0, cy = 0
+        if (bbox) {
+          cx = (bbox.x + bbox.width / 2) * scale
+          cy = (bbox.y + bbox.height / 2) * scale
+        } else if (polygon && polygon.length > 0) {
+          const xs = polygon.map(p => p.x)
+          const ys = polygon.map(p => p.y)
+          cx = ((Math.min(...xs) + Math.max(...xs)) / 2) * scale
+          cy = ((Math.min(...ys) + Math.max(...ys)) / 2) * scale
+        }
+        ctx.save()
+        ctx.font = "bold 12px system-ui, -apple-system, sans-serif"
+        ctx.textAlign = 'center'
+        ctx.textBaseline = 'middle'
+        ctx.fillStyle = 'rgba(15, 23, 42, 0.7)'
+        ctx.beginPath()
+        ctx.arc(cx, cy, 10, 0, Math.PI * 2)
+        ctx.fill()
+        ctx.fillStyle = '#ffffff'
+        ctx.fillText(indexLabel, cx, cy + 0.5)
+        ctx.restore()
+      }
+
+      ctx.restore()
+    }
+
+    // Draw current in-progress selection
+    const m = modeRef.current
+    const { x: sx, y: sy } = startXY.current
+    const { x: ex, y: ey } = currentXY.current
+
+    if (m === 'rect' && (isDragging.current || sx !== ex || sy !== ey)) {
+      const x = Math.min(sx, ex)
+      const y = Math.min(sy, ey)
+      const w = Math.abs(ex - sx)
+      const h2 = Math.abs(ey - sy)
+      if (w > 0 && h2 > 0) {
+        ctx.save()
+        ctx.strokeStyle = '#f97316'
+        ctx.fillStyle = 'rgba(249, 115, 22, 0.3)'
+        ctx.lineWidth = 2
+        ctx.setLineDash([6, 4])
+        ctx.fillRect(x, y, w, h2)
+        ctx.strokeRect(x, y, w, h2)
+        ctx.restore()
+      }
+    } else if (m === 'polygon' && polygonPoints.current.length > 0) {
+      ctx.save()
+      ctx.strokeStyle = '#f97316'
+      ctx.fillStyle = 'rgba(249, 115, 22, 0.3)'
+      ctx.lineWidth = 2
+      ctx.setLineDash([])
+      const pts = polygonPoints.current
+      ctx.beginPath()
+      ctx.moveTo(pts[0].x, pts[0].y)
+      for (let i = 1; i < pts.length; i++) ctx.lineTo(pts[i].x, pts[i].y)
+      if (!isPolygonClosed.current && isDragging.current) ctx.lineTo(ex, ey)
+      if (isPolygonClosed.current) { ctx.closePath(); ctx.fill() }
+      ctx.stroke()
+      for (const p of pts) {
+        ctx.beginPath()
+        ctx.arc(p.x, p.y, 3, 0, Math.PI * 2)
+        ctx.fillStyle = '#ea580c'
+        ctx.fill()
+      }
+      ctx.restore()
+    }
+  }, [])
+
+  useImperativeHandle(ref, () => ({
+    getCurrentSelection(): Selection | null {
+      const scale = displayScale.current
+      if (modeRef.current === 'rect') {
+        const { x: sx, y: sy } = startXY.current
+        const { x: ex, y: ey } = currentXY.current
+        const w = Math.abs(ex - sx)
+        const h = Math.abs(ey - sy)
+        if (w <= 0 || h <= 0) return null
+        return {
+          mode: 'rect',
+          bbox: {
+            x: Math.round(Math.min(sx, ex) / scale),
+            y: Math.round(Math.min(sy, ey) / scale),
+            width: Math.round(w / scale),
+            height: Math.round(h / scale),
+          },
+        }
+      } else {
+        if (!isPolygonClosed.current || polygonPoints.current.length < 3) return null
+        return {
+          mode: 'polygon',
+          polygon: polygonPoints.current.map(p => ({
+            x: Math.round(p.x / scale),
+            y: Math.round(p.y / scale),
+          })),
+        }
+      }
+    },
+    resetSelection() {
+      isDragging.current = false
+      startXY.current = { x: 0, y: 0 }
+      currentXY.current = { x: 0, y: 0 }
+      polygonPoints.current = []
+      isPolygonClosed.current = false
+      onHasSelectionRef.current(false)
+      redraw()
+    },
+  }), [redraw])
+
+  // Reset drawing when mode changes
+  useEffect(() => {
+    modeRef.current = mode
+    isDragging.current = false
+    startXY.current = { x: 0, y: 0 }
+    currentXY.current = { x: 0, y: 0 }
+    polygonPoints.current = []
+    isPolygonClosed.current = false
+    onHasSelectionRef.current(false)
+    redraw()
+  }, [mode, redraw])
+
+  // Load image when src changes
+  useEffect(() => {
+    if (!imageSrc) {
+      imageRef.current = null
+      const canvas = canvasRef.current
+      const ctx = canvas?.getContext('2d')
+      if (canvas && ctx) ctx.clearRect(0, 0, canvas.width, canvas.height)
+      return
+    }
+    const img = new Image()
+    img.onload = () => {
+      const canvas = canvasRef.current
+      if (!canvas) return
+      imageRef.current = img
+      const maxW = (canvas.parentElement?.clientWidth ?? 816) - 16
+      const maxH = window.innerHeight * 0.7
+      const scale = Math.min(maxW / img.width, maxH / img.height, 1)
+      displayScale.current = isFinite(scale) && scale > 0 ? scale : 1
+      canvas.width = img.width * displayScale.current
+      canvas.height = img.height * displayScale.current
+      redraw()
+    }
+    img.onerror = () => console.error('Fehler beim Laden des Bildes:', imageSrc)
+    img.src = imageSrc
+  }, [imageSrc, redraw])
+
+  // Redraw when objects or selection changes
+  useEffect(() => {
+    redraw()
+  }, [objects, selectedObjectId, redraw])
+
+  // Mouse event handlers
+  useEffect(() => {
+    const canvas = canvasRef.current
+    if (!canvas) return
+
+    const getPos = (e: MouseEvent) => {
+      const rect = canvas.getBoundingClientRect()
+      return {
+        x: (e.clientX - rect.left) * (canvas.width / rect.width),
+        y: (e.clientY - rect.top) * (canvas.height / rect.height),
+      }
+    }
+
+    const onMouseDown = (e: MouseEvent) => {
+      if (!imageRef.current) return
+      const { x, y } = getPos(e)
+      if (modeRef.current === 'rect') {
+        startXY.current = { x, y }
+        currentXY.current = { x, y }
+        isDragging.current = true
+      } else {
+        if (isPolygonClosed.current) {
+          polygonPoints.current = []
+          isPolygonClosed.current = false
+          onHasSelectionRef.current(false)
+        }
+        polygonPoints.current.push({ x, y })
+        isDragging.current = true
+      }
+      redraw()
+    }
+
+    const onMouseMove = (e: MouseEvent) => {
+      if (!imageRef.current || !isDragging.current) return
+      currentXY.current = getPos(e)
+      redraw()
+    }
+
+    const onMouseUp = (e: MouseEvent) => {
+      if (!imageRef.current) return
+      isDragging.current = false
+      if (modeRef.current === 'rect') {
+        const w = Math.abs(currentXY.current.x - startXY.current.x)
+        const h = Math.abs(currentXY.current.y - startXY.current.y)
+        onHasSelectionRef.current(w > 0 && h > 0)
+      } else {
+        if (e.detail === 2 && polygonPoints.current.length >= 3) {
+          isPolygonClosed.current = true
+          onHasSelectionRef.current(true)
+        }
+      }
+      redraw()
+    }
+
+    const onMouseLeave = () => {
+      if (isDragging.current) {
+        isDragging.current = false
+        redraw()
+      }
+    }
+
+    canvas.addEventListener('mousedown', onMouseDown)
+    canvas.addEventListener('mousemove', onMouseMove)
+    canvas.addEventListener('mouseup', onMouseUp)
+    canvas.addEventListener('mouseleave', onMouseLeave)
+    return () => {
+      canvas.removeEventListener('mousedown', onMouseDown)
+      canvas.removeEventListener('mousemove', onMouseMove)
+      canvas.removeEventListener('mouseup', onMouseUp)
+      canvas.removeEventListener('mouseleave', onMouseLeave)
+    }
+  }, [redraw])
+
+  return <canvas ref={canvasRef} style={{ display: 'block', maxWidth: '100%', height: 'auto' }} />
+})
--- a/frontend/src/components/ObjectsList.tsx
+++ b/frontend/src/components/ObjectsList.tsx
@@ -0,0 +1,181 @@
+import { useState, useEffect } from 'react'
+import { updateHierarchy, updateObjectMeta, updateParent } from '../api'
+import type { ObjectMeta } from '../types'
+
+interface Props {
+  objects: ObjectMeta[]
+  selectedObjectId: string | null
+  onSelect: (id: string) => void
+  onVisibilityChange?: (id: string, visible: boolean) => void
+  onObjectsChange: (objects: ObjectMeta[]) => void
+  isGeneratePage: boolean
+  onShowDetails?: (obj: ObjectMeta) => void
+  onLoadSentences?: (objId: string) => void
+}
+
+type EditForm = { title_de: string; position_de: string; action_de: string; condition_de: string }
+
+export default function ObjectsList({
+  objects,
+  selectedObjectId,
+  onSelect,
+  onVisibilityChange,
+  onObjectsChange,
+  isGeneratePage,
+  onShowDetails,
+  onLoadSentences,
+}: Props) {
+  const [expandedId, setExpandedId] = useState<string | null>(null)
+  const [editForms, setEditForms] = useState<Record<string, EditForm>>({})
+
+  useEffect(() => {
+    const forms: Record<string, EditForm> = {}
+    for (const obj of objects) {
+      forms[obj.id] = {
+        title_de: obj.title_de || '',
+        position_de: obj.position_de || '',
+        action_de: obj.action_de || '',
+        condition_de: obj.condition_de || '',
+      }
+    }
+    setEditForms(forms)
+  }, [objects])
+
+  const handleHierarchyChange = async (obj: ObjectMeta, value: number) => {
+    onObjectsChange(objects.map(o => o.id === obj.id ? { ...o, hierarchy: value } : o))
+    try { await updateHierarchy(obj.id, value) } catch (e) { console.error(e) }
+  }
+
+  const handleParentChange = async (obj: ObjectMeta, parentId: string | null) => {
+    onObjectsChange(objects.map(o => o.id === obj.id ? { ...o, parent_id: parentId } : o))
+    try { await updateParent(obj.id, parentId) } catch (e) { console.error(e) }
+  }
+
+  const handleSaveMeta = async (obj: ObjectMeta) => {
+    const form = editForms[obj.id]
+    if (!form) return
+    try {
+      await updateObjectMeta(obj.id, form)
+      onObjectsChange(objects.map(o => o.id === obj.id ? { ...o, ...form } : o))
+      setExpandedId(null)
+    } catch (e) { console.error(e) }
+  }
+
+  if (objects.length === 0) {
+    return (
+      <div className="objects-list">
+        <div className="object-item-text">Noch keine Objekte gespeichert.</div>
+      </div>
+    )
+  }
+
+  return (
+    <>
+      <div className="objects-list">
+        {objects.map(obj => (
+          <div
+            key={obj.id}
+            className="object-item"
+            style={obj.id === selectedObjectId ? { borderColor: '#2563eb' } : undefined}
+            onClick={() => {
+              onSelect(obj.id)
+              if (isGeneratePage) {
+                onShowDetails?.(obj)
+                onLoadSentences?.(obj.id)
+              }
+            }}
+          >
+            <div className="object-item-header">
+              {!isGeneratePage && (
+                <input
+                  type="checkbox"
+                  checked={obj.visible !== false}
+                  onClick={e => e.stopPropagation()}
+                  onChange={e => onVisibilityChange?.(obj.id, e.target.checked)}
+                />
+              )}
+              {obj.image_file && (
+                <img
+                  src={`/objects_image/${encodeURIComponent(obj.image_file)}`}
+                  alt={obj.title_de || obj.id}
+                />
+              )}
+              <select
+                className="object-hierarchy-select"
+                value={obj.hierarchy || 1}
+                onClick={e => e.stopPropagation()}
+                onChange={e => handleHierarchyChange(obj, parseInt(e.target.value))}
+              >
+                {[1, 2, 3].map(l => <option key={l} value={l}>{l}</option>)}
+              </select>
+              <select
+                className="object-parent-select"
+                value={obj.parent_id || ''}
+                onClick={e => e.stopPropagation()}
+                onChange={e => handleParentChange(obj, e.target.value || null)}
+              >
+                <option value="">-</option>
+                {objects
+                  .filter(o => o.id !== obj.id)
+                  .map(o => (
+                    <option key={o.id} value={o.id}>{o.index}</option>
+                  ))}
+              </select>
+              <div className="object-item-text">
+                <strong>{obj.title_de || obj.id || 'Ohne Titel'}</strong>
+                {obj.position_de && <span>{obj.position_de}</span>}
+              </div>
+              {!isGeneratePage && (
+                <button
+                  type="button"
+                  className="object-icon-button"
+                  onClick={e => {
+                    e.stopPropagation()
+                    setExpandedId(expandedId === obj.id ? null : obj.id)
+                  }}
+                >
+                  📝
+                </button>
+              )}
+            </div>
+
+            {!isGeneratePage && expandedId === obj.id && (
+              <div className="object-item-details visible">
+                {(['title_de', 'position_de', 'action_de', 'condition_de'] as const).map(key => (
+                  <div key={key}>
+                    <label style={{ fontSize: '0.75rem' }}>{key}</label>
+                    <input
+                      type="text"
+                      value={editForms[obj.id]?.[key] || ''}
+                      onClick={e => e.stopPropagation()}
+                      onChange={e =>
+                        setEditForms(f => ({
+                          ...f,
+                          [obj.id]: { ...f[obj.id], [key]: e.target.value },
+                        }))
+                      }
+                    />
+                  </div>
+                ))}
+                <button
+                  type="button"
+                  className="object-icon-button"
+                  onClick={e => { e.stopPropagation(); handleSaveMeta(obj) }}
+                >
+                  💾
+                </button>
+              </div>
+            )}
+          </div>
+        ))}
+      </div>
+      <div className="objects-tags">
+        {objects.map(obj => (
+          <span key={obj.id} className="object-tag">
+            {obj.title_de || obj.id || 'Ohne Titel'}
+          </span>
+        ))}
+      </div>
+    </>
+  )
+}
--- a/frontend/src/components/SentencesList.tsx
+++ b/frontend/src/components/SentencesList.tsx
@@ -0,0 +1,34 @@
+import type { Sentence } from '../types'
+
+interface Props {
+  sentences: Sentence[]
+}
+
+export default function SentencesList({ sentences }: Props) {
+  if (sentences.length === 0) {
+    return (
+      <div className="sentences-list">
+        <div className="sentence-item-empty">Noch keine Sätze vorhanden.</div>
+      </div>
+    )
+  }
+
+  return (
+    <div className="sentences-list">
+      {[...sentences].reverse().map((s, i) => (
+        <div className="sentence-item" key={i}>
+          <div style={{ fontWeight: 600, color: '#1e40af', marginBottom: 4, fontSize: '0.85rem' }}>
+            Einfach:
+          </div>
+          <div className="sentence-item-question">{s.question_simple_en}</div>
+          <div className="sentence-item-answer">{s.answer_simple_en}</div>
+          <div style={{ fontWeight: 600, color: '#1e40af', marginTop: 8, marginBottom: 4, fontSize: '0.85rem' }}>
+            Fortgeschritten:
+          </div>
+          <div className="sentence-item-question">{s.question_advanced_en}</div>
+          <div className="sentence-item-answer">{s.answer_advanced_en}</div>
+        </div>
+      ))}
+    </div>
+  )
+}
--- a/frontend/src/index.css
+++ b/frontend/src/index.css
@@ -0,0 +1,382 @@
+body {
+  font-family: system-ui, -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;
+  margin: 0;
+  padding: 0;
+  background: #f5f7fb;
+  color: #222;
+}
+
+.container {
+  max-width: 1180px;
+  margin: 24px auto;
+  padding: 16px 20px 32px;
+  background: #ffffff;
+  border-radius: 12px;
+  box-shadow: 0 10px 30px rgba(15, 23, 42, 0.08);
+}
+
+h1 {
+  margin-top: 0;
+  margin-bottom: 16px;
+  font-size: 1.6rem;
+}
+
+label {
+  font-weight: 500;
+}
+
+.panel {
+  margin: 12px 0;
+  display: flex;
+  align-items: center;
+  gap: 8px;
+  flex-wrap: wrap;
+}
+
+.image-nav {
+  justify-content: space-between;
+  gap: 8px;
+}
+
+.image-nav button {
+  padding: 4px 10px;
+  border-radius: 999px;
+}
+
+.image-nav-left {
+  display: flex;
+  align-items: center;
+  gap: 8px;
+}
+
+.page-switch select {
+  padding: 4px 8px;
+  border-radius: 999px;
+  border: 1px solid #cbd5f0;
+  background: #f9fbff;
+  font-size: 0.9rem;
+  min-width: unset;
+}
+
+.main-layout {
+  display: flex;
+  gap: 16px;
+  align-items: flex-start;
+}
+
+.objects-pane {
+  flex: 0 0 260px;
+}
+
+.left-pane {
+  flex: 1 1 auto;
+}
+
+.right-pane {
+  flex: 0 0 280px;
+  display: flex;
+  flex-direction: column;
+  gap: 12px;
+}
+
+.sentences-pane {
+  flex: 0 0 300px;
+  display: flex;
+  flex-direction: column;
+  gap: 12px;
+}
+
+.mode-option {
+  display: inline-flex;
+  align-items: center;
+  gap: 4px;
+  font-weight: 400;
+}
+
+select {
+  min-width: 220px;
+  padding: 6px 10px;
+  border-radius: 8px;
+  border: 1px solid #cbd5f0;
+  background: #f9fbff;
+}
+
+.canvas-wrapper {
+  border-radius: 12px;
+  border: 1px solid #d4ddf5;
+  background: #f1f5ff;
+  overflow: visible;
+  padding: 8px;
+}
+
+canvas {
+  display: block;
+  max-width: 100%;
+  height: auto;
+}
+
+button {
+  padding: 8px 14px;
+  border-radius: 999px;
+  border: none;
+  background: #2563eb;
+  color: white;
+  font-weight: 500;
+  cursor: pointer;
+  box-shadow: 0 4px 10px rgba(37, 99, 235, 0.4);
+  transition: background 0.15s ease, box-shadow 0.15s ease, transform 0.1s ease;
+}
+
+button:disabled {
+  background: #9ca3af;
+  cursor: not-allowed;
+  box-shadow: none;
+}
+
+button:not(:disabled):hover {
+  background: #1d4ed8;
+  box-shadow: 0 6px 18px rgba(37, 99, 235, 0.5);
+  transform: translateY(-1px);
+}
+
+.status {
+  font-size: 0.9rem;
+}
+
+.status.ok {
+  color: #16a34a;
+}
+
+.status.error {
+  color: #dc2626;
+}
+
+.sidebar-section {
+  border-radius: 10px;
+  border: 1px solid #e5e7eb;
+  padding: 10px 12px;
+  background: #f9fafb;
+}
+
+.sidebar-section h2 {
+  font-size: 1rem;
+  margin: 0 0 6px;
+}
+
+.sidebar-row {
+  display: flex;
+  flex-direction: column;
+  gap: 4px;
+  margin-top: 6px;
+}
+
+.sidebar-row input[type="text"] {
+  padding: 6px 8px;
+  border-radius: 6px;
+  border: 1px solid #d1d5db;
+  font-size: 0.9rem;
+}
+
+.detail-value {
+  padding: 4px 6px;
+  border-radius: 6px;
+  border: 1px solid #e5e7eb;
+  background: #f9fafb;
+  font-size: 0.9rem;
+  min-height: 24px;
+}
+
+.objects-list {
+  display: flex;
+  flex-direction: column;
+  gap: 8px;
+  max-height: 260px;
+  overflow-y: auto;
+  padding-right: 4px;
+}
+
+.objects-tags {
+  margin-top: 8px;
+  display: flex;
+  flex-wrap: wrap;
+  gap: 4px;
+}
+
+.object-tag {
+  padding: 2px 6px;
+  border-radius: 999px;
+  background: #e5e7eb;
+  font-size: 0.75rem;
+  white-space: nowrap;
+}
+
+.object-item {
+  display: flex;
+  flex-direction: column;
+  gap: 4px;
+  padding: 4px 6px;
+  border-radius: 8px;
+  border: 1px solid #e5e7eb;
+  background: #ffffff;
+  cursor: pointer;
+}
+
+.object-item-header {
+  display: flex;
+  align-items: center;
+  gap: 4px;
+  flex-wrap: nowrap;
+}
+
+.object-hierarchy-select {
+  width: 40px;
+  min-width: 0;
+  padding: 2px 3px;
+  border-radius: 6px;
+  border: 1px solid #d1d5db;
+  font-size: 0.8rem;
+  background: #f9fafb;
+}
+
+.object-parent-select {
+  width: 50px;
+  min-width: 0;
+  padding: 2px 3px;
+  border-radius: 6px;
+  border: 1px solid #d1d5db;
+  font-size: 0.8rem;
+  background: #eef2ff;
+}
+
+.object-item img {
+  width: 40px;
+  height: 40px;
+  object-fit: cover;
+  border-radius: 6px;
+  border: 1px solid #e5e7eb;
+}
+
+.object-item-text {
+  display: flex;
+  flex-direction: column;
+  font-size: 0.8rem;
+  flex: 1;
+  min-width: 0;
+}
+
+.object-item-text strong {
+  font-size: 0.85rem;
+}
+
+.object-item-details {
+  padding-left: 24px;
+  display: none;
+  flex-direction: column;
+  gap: 4px;
+  margin-top: 4px;
+}
+
+.object-item-details.visible {
+  display: flex;
+}
+
+.object-item-details input[type="text"] {
+  padding: 4px 6px;
+  border-radius: 6px;
+  border: 1px solid #d1d5db;
+  font-size: 0.8rem;
+}
+
+.object-item-details label {
+  font-size: 0.75rem;
+}
+
+.object-icon-button {
+  padding: 2px 6px;
+  border-radius: 6px;
+  font-size: 0.85rem;
+  box-shadow: none;
+  background: #e5e7eb;
+  color: #374151;
+  flex-shrink: 0;
+}
+
+.object-icon-button:not(:disabled):hover {
+  background: #d1d5db;
+  box-shadow: none;
+  transform: none;
+}
+
+.sentences-list {
+  display: flex;
+  flex-direction: column;
+  gap: 12px;
+  max-height: 70vh;
+  overflow-y: auto;
+  padding: 4px;
+}
+
+.sentence-item {
+  padding: 12px;
+  background: #f9fbff;
+  border: 1px solid #d4ddf5;
+  border-radius: 8px;
+  display: flex;
+  flex-direction: column;
+  gap: 8px;
+}
+
+.sentence-item-question {
+  font-weight: 600;
+  color: #1e40af;
+  font-size: 0.95rem;
+}
+
+.sentence-item-answer {
+  color: #475569;
+  font-size: 0.9rem;
+  padding-left: 12px;
+  border-left: 2px solid #cbd5f0;
+}
+
+.sentence-item-empty {
+  padding: 24px;
+  text-align: center;
+  color: #94a3b8;
+  font-style: italic;
+}
+
+.selections-list {
+  display: flex;
+  flex-direction: column;
+  gap: 8px;
+  max-height: 200px;
+  overflow-y: auto;
+  padding: 8px;
+  background: #f9fbff;
+  border: 1px solid #d4ddf5;
+  border-radius: 8px;
+  margin-bottom: 8px;
+}
+
+.selection-item {
+  padding: 8px;
+  background: #ffffff;
+  border: 1px solid #e5e7eb;
+  border-radius: 6px;
+  font-size: 0.85rem;
+  color: #475569;
+}
+
+.selection-item strong {
+  color: #1e40af;
+  font-weight: 600;
+}
+
+.selections-empty {
+  padding: 16px;
+  text-align: center;
+  color: #94a3b8;
+  font-style: italic;
+  font-size: 0.85rem;
+}
--- a/frontend/src/main.tsx
+++ b/frontend/src/main.tsx
@@ -0,0 +1,13 @@
+import { StrictMode } from 'react'
+import { createRoot } from 'react-dom/client'
+import { BrowserRouter } from 'react-router-dom'
+import App from './App'
+import './index.css'
+
+createRoot(document.getElementById('root')!).render(
+  <StrictMode>
+    <BrowserRouter>
+      <App />
+    </BrowserRouter>
+  </StrictMode>
+)
--- a/frontend/src/pages/DrawIt.tsx
+++ b/frontend/src/pages/DrawIt.tsx
@@ -0,0 +1,298 @@
+import { useState, useEffect, useCallback, useRef } from 'react'
+import { useNavigate } from 'react-router-dom'
+import DrawCanvas, { type DrawCanvasHandle } from '../components/DrawCanvas'
+import ObjectsList from '../components/ObjectsList'
+import { getImages, getObjects, cropImage, saveImage } from '../api'
+import type { ObjectMeta, Selection } from '../types'
+
+const FIELD_LABELS: Record<string, string> = {
+  title_de: 'Titel / title_de',
+  position_de: 'Position / position_de',
+  action_de: 'Status (sitzt/schwimmt/segelt) / action_de',
+  condition_de: 'Zustand (alt/jung/rostig) / condition_de',
+}
+
+const FIELD_PLACEHOLDERS: Record<string, string> = {
+  action_de: 'z.B. sitzt',
+  condition_de: 'z.B. rostig',
+}
+
+type FormKey = 'title_de' | 'position_de' | 'action_de' | 'condition_de'
+
+export default function DrawIt() {
+  const navigate = useNavigate()
+
+  const [imageList, setImageList] = useState<string[]>([])
+  const [currentIndex, setCurrentIndex] = useState(-1)
+  const [objects, setObjects] = useState<ObjectMeta[]>([])
+  const [currentSelections, setCurrentSelections] = useState<Selection[]>([])
+  const [status, setStatus] = useState('')
+  const [statusError, setStatusError] = useState(false)
+  const [mode, setMode] = useState<'rect' | 'polygon'>('rect')
+  const [hasSelection, setHasSelection] = useState(false)
+  const [selectedObjectId, setSelectedObjectId] = useState<string | null>(null)
+  const [form, setForm] = useState<Record<FormKey, string>>({
+    title_de: '',
+    position_de: '',
+    action_de: '',
+    condition_de: '',
+  })
+
+  const canvasRef = useRef<DrawCanvasHandle>(null)
+  const currentFilename = currentIndex >= 0 && currentIndex < imageList.length
+    ? imageList[currentIndex]
+    : null
+
+  useEffect(() => {
+    getImages('draw')
+      .then(imgs => {
+        setImageList(imgs)
+        setCurrentIndex(imgs.length - 1)
+      })
+      .catch(console.error)
+  }, [])
+
+  useEffect(() => {
+    if (!currentFilename) {
+      setObjects([])
+      setSelectedObjectId(null)
+      return
+    }
+    getObjects(currentFilename)
+      .then(objs => {
+        setObjects(objs.map(o => ({ ...o, visible: true })))
+        setSelectedObjectId(objs[0]?.id ?? null)
+      })
+      .catch(console.error)
+  }, [currentFilename])
+
+  const handleHasSelection = useCallback((has: boolean) => setHasSelection(has), [])
+
+  const showStatus = (msg: string, isError = false) => {
+    setStatus(msg)
+    setStatusError(isError)
+  }
+
+  const addSelection = () => {
+    const sel = canvasRef.current?.getCurrentSelection()
+    if (!sel) {
+      showStatus('Bitte zuerst einen Bereich auswählen.', true)
+      return
+    }
+    setCurrentSelections(prev => {
+      const next = [...prev, sel]
+      showStatus(`Auswahl ${next.length} hinzugefügt.`)
+      return next
+    })
+    canvasRef.current?.resetSelection()
+    setHasSelection(false)
+  }
+
+  const saveObject = async () => {
+    if (!currentFilename || currentSelections.length === 0) return
+    try {
+      showStatus('Speichere Objekt...')
+      const result = await cropImage({
+        filename: currentFilename,
+        selections: currentSelections.map((sel, idx) => ({
+          number: idx + 1,
+          mode: sel.mode,
+          bbox: sel.bbox ?? null,
+          polygon: sel.polygon ?? null,
+        })),
+        ...form,
+      })
+      showStatus(`Gespeichert – ID: ${result.id} (${currentSelections.length} Auswahl(en))`)
+      setCurrentSelections([])
+      setForm({ title_de: '', position_de: '', action_de: '', condition_de: '' })
+      const objs = await getObjects(currentFilename)
+      setObjects(objs.map(o => ({ ...o, visible: true })))
+    } catch (e) {
+      showStatus(e instanceof Error ? e.message : 'Fehler beim Speichern.', true)
+    }
+  }
+
+  const handleSaveImage = async () => {
+    if (!currentFilename) return
+    try {
+      showStatus('Bild wird gespeichert...')
+      await saveImage(currentFilename)
+      const imgs = await getImages('draw')
+      setImageList(imgs)
+      setCurrentIndex(imgs.length - 1)
+      showStatus('Bild gespeichert.')
+    } catch (e) {
+      showStatus(e instanceof Error ? e.message : 'Fehler.', true)
+    }
+  }
+
+  return (
+    <div className="container">
+      <h1>DrawIt</h1>
+
+      <div className="panel image-nav">
+        <div className="image-nav-left">
+          <button
+            onClick={() => setCurrentIndex(i => i - 1)}
+            disabled={currentIndex <= 0}
+          >
+            ⬅️
+          </button>
+          <span>Bild: <code>{currentFilename || '–'}</code></span>
+          <button
+            onClick={() => setCurrentIndex(i => i + 1)}
+            disabled={currentIndex >= imageList.length - 1}
+          >
+            ➡️
+          </button>
+          <button onClick={handleSaveImage} disabled={!currentFilename}>💾</button>
+        </div>
+        <div className="page-switch">
+          <select value="/draw" onChange={e => navigate(e.target.value)}>
+            <option value="/draw">DrawIt</option>
+            <option value="/generate">GenerateIt</option>
+          </select>
+        </div>
+      </div>
+
+      <div className="main-layout">
+        {/* Left: Objects */}
+        <div className="objects-pane sidebar-section">
+          <h2>Objekte zu diesem Bild</h2>
+          <ObjectsList
+            objects={objects}
+            selectedObjectId={selectedObjectId}
+            onSelect={setSelectedObjectId}
+            onVisibilityChange={(id, visible) =>
+              setObjects(prev => prev.map(o => o.id === id ? { ...o, visible } : o))
+            }
+            onObjectsChange={setObjects}
+            isGeneratePage={false}
+          />
+        </div>
+
+        {/* Center: Canvas */}
+        <div className="left-pane">
+          <div className="canvas-wrapper">
+            <DrawCanvas
+              ref={canvasRef}
+              imageSrc={
+                currentFilename
+                  ? `/pictures/${encodeURIComponent(currentFilename)}`
+                  : null
+              }
+              objects={objects}
+              selectedObjectId={selectedObjectId}
+              mode={mode}
+              onHasSelection={handleHasSelection}
+            />
+          </div>
+        </div>
+
+        {/* Right: Controls */}
+        <div className="right-pane">
+          <div className="sidebar-section">
+            <h2>Auswahl</h2>
+            <div className="sidebar-row">
+              <span>Auswahl-Typ (Interface / Backend):</span>
+            </div>
+            <div className="sidebar-row">
+              <label className="mode-option">
+                <input
+                  type="radio"
+                  name="mode"
+                  value="rect"
+                  checked={mode === 'rect'}
+                  onChange={() => setMode('rect')}
+                />
+                Rechteck / <code>BBox</code>
+              </label>
+            </div>
+            <div className="sidebar-row">
+              <label className="mode-option">
+                <input
+                  type="radio"
+                  name="mode"
+                  value="polygon"
+                  checked={mode === 'polygon'}
+                  onChange={() => setMode('polygon')}
+                />
+                Polygon / <code>polygon</code>
+              </label>
+            </div>
+            <div className="sidebar-row">
+              <button type="button" onClick={() => canvasRef.current?.resetSelection()}>
+                Auswahl zurücksetzen
+              </button>
+            </div>
+          </div>
+
+          <div className="sidebar-section">
+            <h2>Metadaten</h2>
+            {(['title_de', 'position_de', 'action_de', 'condition_de'] as FormKey[]).map(key => (
+              <div className="sidebar-row" key={key}>
+                <label htmlFor={key}>{FIELD_LABELS[key]}</label>
+                <input
+                  id={key}
+                  type="text"
+                  value={form[key]}
+                  placeholder={FIELD_PLACEHOLDERS[key] || ''}
+                  onChange={e => setForm(f => ({ ...f, [key]: e.target.value }))}
+                />
+              </div>
+            ))}
+          </div>
+
+          <div className="sidebar-section">
+            <h2>Auswahlen</h2>
+            <div className="selections-list">
+              {currentSelections.length === 0 ? (
+                <div className="selections-empty">Noch keine Auswahlen hinzugefügt.</div>
+              ) : (
+                currentSelections.map((sel, i) => (
+                  <div className="selection-item" key={i}>
+                    <strong>Auswahl {i + 1}</strong> ({sel.mode === 'rect' ? 'Rechteck' : 'Polygon'}):
+                    {sel.mode === 'rect' && sel.bbox
+                      ? ` x=${sel.bbox.x}, y=${sel.bbox.y}, w=${sel.bbox.width}, h=${sel.bbox.height}`
+                      : ` ${sel.polygon?.length ?? 0} Punkte`}
+                  </div>
+                ))
+              )}
+            </div>
+            <div className="sidebar-row">
+              <button
+                onClick={addSelection}
+                disabled={!hasSelection || !currentFilename}
+              >
+                ➕ Auswahl hinzufügen
+              </button>
+            </div>
+            <div className="sidebar-row">
+              <button
+                onClick={saveObject}
+                disabled={!currentFilename || currentSelections.length === 0}
+              >
+                💾 Objekt speichern
+              </button>
+            </div>
+            <div className="sidebar-row">
+              <button
+                type="button"
+                onClick={() => {
+                  setCurrentSelections([])
+                  canvasRef.current?.resetSelection()
+                  showStatus('Alle Auswahlen gelöscht.')
+                }}
+              >
+                🗑️ Alle Auswahlen löschen
+              </button>
+            </div>
+            {status && (
+              <span className={`status ${statusError ? 'error' : 'ok'}`}>{status}</span>
+            )}
+          </div>
+        </div>
+      </div>
+    </div>
+  )
+}
--- a/frontend/src/pages/GenerateIt.tsx
+++ b/frontend/src/pages/GenerateIt.tsx
@@ -0,0 +1,166 @@
+import { useState, useEffect } from 'react'
+import { useNavigate } from 'react-router-dom'
+import ObjectsList from '../components/ObjectsList'
+import DetailsPanel from '../components/DetailsPanel'
+import SentencesList from '../components/SentencesList'
+import { getImages, getObjects, generateDetails, generateSentence, getSentences } from '../api'
+import type { ObjectMeta, Sentence } from '../types'
+
+export default function GenerateIt() {
+  const navigate = useNavigate()
+
+  const [imageList, setImageList] = useState<string[]>([])
+  const [currentIndex, setCurrentIndex] = useState(-1)
+  const [objects, setObjects] = useState<ObjectMeta[]>([])
+  const [selectedObj, setSelectedObj] = useState<ObjectMeta | null>(null)
+  const [sentences, setSentences] = useState<Sentence[]>([])
+  const [isGeneratingDetails, setIsGeneratingDetails] = useState(false)
+  const [isGeneratingSentence, setIsGeneratingSentence] = useState(false)
+
+  const currentFilename =
+    currentIndex >= 0 && currentIndex < imageList.length ? imageList[currentIndex] : null
+
+  useEffect(() => {
+    getImages('generate')
+      .then(imgs => {
+        setImageList(imgs)
+        setCurrentIndex(imgs.length - 1)
+      })
+      .catch(console.error)
+  }, [])
+
+  useEffect(() => {
+    if (!currentFilename) {
+      setObjects([])
+      setSelectedObj(null)
+      setSentences([])
+      return
+    }
+    getObjects(currentFilename)
+      .then(objs => {
+        setObjects(objs)
+        if (objs.length > 0) {
+          setSelectedObj(objs[0])
+          loadSentences(objs[0].id)
+        } else {
+          setSelectedObj(null)
+          setSentences([])
+        }
+      })
+      .catch(console.error)
+  }, [currentFilename])
+
+  const loadSentences = async (objId: string) => {
+    try {
+      const s = await getSentences(objId)
+      setSentences(s)
+    } catch (e) {
+      console.error(e)
+    }
+  }
+
+  const handleGenerateDetails = async () => {
+    const target = selectedObj ?? objects[0]
+    if (!target) return
+    setIsGeneratingDetails(true)
+    try {
+      const data = await generateDetails(target.id)
+      const updated = { ...target, ...data }
+      setSelectedObj(updated)
+      setObjects(prev => prev.map(o => o.id === target.id ? updated : o))
+    } catch (e) {
+      alert(e instanceof Error ? e.message : 'Fehler bei KI-Details')
+    } finally {
+      setIsGeneratingDetails(false)
+    }
+  }
+
+  const handleGenerateSentence = async () => {
+    const target = selectedObj ?? objects[0]
+    if (!target) return
+    setIsGeneratingSentence(true)
+    try {
+      const data = await generateSentence(target.id)
+      setSentences(prev => [...prev, data.sentence])
+      setSelectedObj(prev => prev ? { ...prev, latest_sentence: data.sentence } : prev)
+    } catch (e) {
+      alert(e instanceof Error ? e.message : 'Fehler bei KI-Sentence')
+    } finally {
+      setIsGeneratingSentence(false)
+    }
+  }
+
+  return (
+    <div className="container">
+      <h1>GenerateIt</h1>
+
+      <div className="panel image-nav">
+        <div className="image-nav-left">
+          <button
+            onClick={() => setCurrentIndex(i => i - 1)}
+            disabled={currentIndex <= 0}
+          >
+            ⬅️
+          </button>
+          <span>Bild: <code>{currentFilename || '–'}</code></span>
+          <button
+            onClick={() => setCurrentIndex(i => i + 1)}
+            disabled={currentIndex >= imageList.length - 1}
+          >
+            ➡️
+          </button>
+          <button
+            onClick={handleGenerateDetails}
+            disabled={isGeneratingDetails || !selectedObj}
+          >
+            {isGeneratingDetails ? '⏳ KI-Details...' : '✨ KI-Details'}
+          </button>
+          <button
+            onClick={handleGenerateSentence}
+            disabled={isGeneratingSentence || !selectedObj}
+          >
+            {isGeneratingSentence ? '⏳ KI-Sentence...' : '💬 KI-Sentence'}
+          </button>
+        </div>
+        <div className="page-switch">
+          <select value="/generate" onChange={e => navigate(e.target.value)}>
+            <option value="/draw">DrawIt</option>
+            <option value="/generate">GenerateIt</option>
+          </select>
+        </div>
+      </div>
+
+      <div className="main-layout">
+        <div className="objects-pane sidebar-section">
+          <h2>Objekte zu diesem Bild</h2>
+          <ObjectsList
+            objects={objects}
+            selectedObjectId={selectedObj?.id ?? null}
+            onSelect={id => {
+              const obj = objects.find(o => o.id === id)
+              if (obj) {
+                setSelectedObj(obj)
+                loadSentences(obj.id)
+              }
+            }}
+            onObjectsChange={setObjects}
+            isGeneratePage={true}
+            onShowDetails={obj => setSelectedObj(obj)}
+            onLoadSentences={loadSentences}
+          />
+        </div>
+
+        <div className="right-pane">
+          <DetailsPanel obj={selectedObj} objects={objects} sentences={sentences} />
+        </div>
+
+        <div className="sentences-pane">
+          <div className="sidebar-section">
+            <h2>Alle Sätze</h2>
+            <SentencesList sentences={sentences} />
+          </div>
+        </div>
+      </div>
+    </div>
+  )
+}
--- a/frontend/src/types.ts
+++ b/frontend/src/types.ts
@@ -0,0 +1,53 @@
+export interface BBox {
+  x: number
+  y: number
+  width: number
+  height: number
+}
+
+export interface Point {
+  x: number
+  y: number
+}
+
+export interface Selection {
+  mode: 'rect' | 'polygon'
+  bbox?: BBox
+  polygon?: Point[]
+}
+
+export interface Sentence {
+  object_id: string
+  created_at: string
+  question_simple_en: string
+  answer_simple_en: string
+  question_advanced_en: string
+  answer_advanced_en: string
+}
+
+export interface ObjectMeta {
+  id: string
+  image_file: string
+  title_de: string
+  position_de: string
+  action_de: string
+  condition_de: string
+  label_en?: string
+  label_de?: string
+  label_se?: string
+  color_en?: string
+  adjective_en?: string
+  action_verb_en?: string
+  preposition_en?: string
+  relative_position_en?: string
+  season_en?: string
+  mode: 'rect' | 'polygon'
+  bbox?: BBox
+  polygon?: Point[]
+  hierarchy: number
+  parent_id?: string | null
+  created_at: string
+  index?: number
+  visible?: boolean
+  latest_sentence?: Sentence
+}
--- a/frontend/tsconfig.json
+++ b/frontend/tsconfig.json
@@ -0,0 +1,18 @@
+{
+  "compilerOptions": {
+    "target": "ES2020",
+    "useDefineForClassFields": true,
+    "lib": ["ES2020", "DOM", "DOM.Iterable"],
+    "module": "ESNext",
+    "skipLibCheck": true,
+    "moduleResolution": "bundler",
+    "allowImportingTsExtensions": true,
+    "resolveJsonModule": true,
+    "isolatedModules": true,
+    "noEmit": true,
+    "jsx": "react-jsx",
+    "strict": true
+  },
+  "include": ["src"],
+  "references": [{ "path": "./tsconfig.node.json" }]
+}
--- a/frontend/tsconfig.node.json
+++ b/frontend/tsconfig.node.json
@@ -0,0 +1,10 @@
+{
+  "compilerOptions": {
+    "composite": true,
+    "skipLibCheck": true,
+    "module": "ESNext",
+    "moduleResolution": "bundler",
+    "allowSyntheticDefaultImports": true
+  },
+  "include": ["vite.config.ts"]
+}
--- a/frontend/vite.config.ts
+++ b/frontend/vite.config.ts
@@ -0,0 +1,18 @@
+import { defineConfig } from 'vite'
+import react from '@vitejs/plugin-react'
+
+export default defineConfig({
+  plugins: [react()],
+  server: {
+    port: 3000,
+    proxy: {
+      '/api': 'http://localhost:8000',
+      '/pictures': 'http://localhost:8000',
+      '/objects_image': 'http://localhost:8000',
+    },
+  },
+  build: {
+    outDir: '../static/react',
+    emptyOutDir: true,
+  },
+})
--- a/prompts/create_details.txt
+++ b/prompts/create_details.txt
@@ -0,0 +1,39 @@
+Du bist eine hilfreiche Bild-KI auf Basis von Llama 3.1.
+
+Du erhältst:
+- einen BILDAUSSCHNITT eines einzelnen Objektes (z.B. ausgeschnittenes Boot, Mensch, Tier, Gebäude, usw.)
+- bereits vorhandene Metadaten zum Objekt in DEUTSCH.
+
+Deine Aufgabe:
+- Analysiere ausschließlich dieses eine Objekt im Bildausschnitt.
+- Nutze die deutschen Metadaten nur als Zusatzkontext, falls sinnvoll.
+- Antworte **immer** mit GENAU EINEM JSON-OBJEKT, ohne erklärenden Text davor oder dahinter.
+- Verwende **englische** Begriffe in den Werten, außer dort, wo explizit etwas anderes verlangt ist.
+
+Rules for fields:
+1. label_en: Single noun, singular (e.g., "cloud", "lighthouse").
+2. label_de: German translation, singular, lowercase (e.g., "wolke").
+3. label_se: Swedish translation, singular, lowercase (e.g., "moln").
+4. color_en: Main color as a single word (e.g., "white", "grey").
+5. adjective_en: One descriptive adjective about condition/appearance (e.g., "fluffy", "dark", "weathered").
+6. action_verb_en: One verb ending in -ing describing the state (e.g., "floating", "shining", "standing").
+7. preposition_en: The most fitting preposition to describe its location (e.g., "in", "on", "above", "behind").
+8. relative_position_en: The object it is positioned relative to (e.g., "sky", "water", "hill").
+9. season_en: The most likely season (e.g., "summer", "autumn").
+
+Output format:
+{
+  "label_en": "...",
+  "label_de": "...",
+  "label_se": "...",
+  "color_en": "...",
+  "adjective_en": "...",
+  "action_verb_en": "...",
+  "preposition_en": "...",
+  "relative_position_en": "...",
+  "season_en": "..."
+}
+
+WICHTIG:
+- Gib **nur** dieses JSON-Objekt zurück.
+- Keine Erklärungen, keine Kommentare, kein Markdown.
--- a/prompts/create_sentence.txt
+++ b/prompts/create_sentence.txt
@@ -0,0 +1,44 @@
+You are an expert ESL teacher and dialogue writer.
+
+Your goal: Create **TWO** short, natural **English question and answer pairs** that a language teacher can use to talk about ONE object in an image.
+
+You will receive, in the user message:
+- JSON with details about the object (fields like label_en, label_de, color_en, adjective_en, action_verb_en, preposition_en, relative_position_en, season_en, etc.)
+- JSON list of previous sentences for this object (each with question_simple_en, answer_simple_en, question_advanced_en, answer_advanced_en).
+
+### Task
+1. Read the object details and understand what the object is, what it looks like, what it is doing and where it is.
+2. Read the previous sentences and **avoid repeating the same question or answer meaning**.
+3. Create **TWO** question and answer pairs in English:
+   - **SIMPLE** (for children): Very simple language, A1 level. Short words, no complex grammar. Avoid "like", "tam tam" or similar filler words. Direct and clear.
+   - **ADVANCED** (for learners): Natural, clear English, about A2–B1 level. More descriptive, natural flow.
+
+### Style examples
+
+**Simple (for children):**
+- Q: "Can you see the boat?"
+  A: "great, you can see the boat."
+- Q: "Where is the bird?"
+  A: "The bird is in the sky."
+
+**Advanced (for learners):**
+- Q: "Can you see the seagull flying in the sky?"
+  A: "Yes, you can see a seagull flying in the sky."
+- Q: "Where is the red boat?"
+  A: "The red boat is on the water near the lighthouse."
+
+### Output format
+Return **exactly one** JSON object, with **no extra text** before or after:
+{
+  "question_simple_en": "...",
+  "answer_simple_en": "...",
+  "question_advanced_en": "...",
+  "answer_advanced_en": "..."
+}
+
+Important:
+- Do NOT add any other fields.
+- Do NOT wrap this object in a list.
+- Do NOT explain your reasoning.
+- Simple sentences must be VERY simple - suitable for young children learning English.
+- Never use "what" or "why" 
--- a/requirements.txt
+++ b/requirements.txt
@@ -0,0 +1,6 @@
+Flask==3.0.3
+Flask-Cors==4.0.1
+gunicorn==22.0.0
+Pillow==11.0.0
+ollama==0.3.0
+
--- a/static/script.js
+++ b/static/script.js
--- a/static/style.css
+++ b/static/style.css
@@ -0,0 +1,369 @@
+body {
+  font-family: system-ui, -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;
+  margin: 0;
+  padding: 0;
+  background: #f5f7fb;
+  color: #222;
+}
+
+.container {
+  max-width: 1180px;
+  margin: 24px auto;
+  padding: 16px 20px 32px;
+  background: #ffffff;
+  border-radius: 12px;
+  box-shadow: 0 10px 30px rgba(15, 23, 42, 0.08);
+}
+
+h1 {
+  margin-top: 0;
+  margin-bottom: 16px;
+  font-size: 1.6rem;
+}
+
+label {
+  font-weight: 500;
+}
+
+.panel {
+  margin: 12px 0;
+  display: flex;
+  align-items: center;
+  gap: 8px;
+  flex-wrap: wrap;
+}
+
+.image-nav {
+  justify-content: space-between;
+  gap: 8px;
+}
+
+.image-nav button {
+  padding: 4px 10px;
+  border-radius: 999px;
+}
+
+.image-nav-left {
+  display: flex;
+  align-items: center;
+  gap: 8px;
+}
+
+.page-switch select {
+  padding: 4px 8px;
+  border-radius: 999px;
+  border: 1px solid #cbd5f0;
+  background: #f9fbff;
+  font-size: 0.9rem;
+}
+
+.main-layout {
+  display: flex;
+  gap: 16px;
+  align-items: flex-start;
+}
+
+.objects-pane {
+  flex: 0 0 260px;
+}
+
+.left-pane {
+  flex: 1 1 auto;
+}
+
+.right-pane {
+  flex: 0 0 280px;
+  display: flex;
+  flex-direction: column;
+  gap: 12px;
+}
+
+.sentences-pane {
+  flex: 0 0 300px;
+  display: flex;
+  flex-direction: column;
+  gap: 12px;
+}
+
+.mode-option {
+  display: inline-flex;
+  align-items: center;
+  gap: 4px;
+  font-weight: 400;
+}
+
+select {
+  min-width: 220px;
+  padding: 6px 10px;
+  border-radius: 8px;
+  border: 1px solid #cbd5f0;
+  background: #f9fbff;
+}
+
+.canvas-wrapper {
+  border-radius: 12px;
+  border: 1px solid #d4ddf5;
+  background: #f1f5ff;
+  overflow: visible;
+  padding: 8px;
+}
+
+canvas {
+  display: block;
+  max-width: 100%;
+  height: auto;
+}
+
+button {
+  padding: 8px 14px;
+  border-radius: 999px;
+  border: none;
+  background: #2563eb;
+  color: white;
+  font-weight: 500;
+  cursor: pointer;
+  box-shadow: 0 4px 10px rgba(37, 99, 235, 0.4);
+  transition: background 0.15s ease, box-shadow 0.15s ease, transform 0.1s ease;
+}
+
+button:disabled {
+  background: #9ca3af;
+  cursor: not-allowed;
+  box-shadow: none;
+}
+
+button:not(:disabled):hover {
+  background: #1d4ed8;
+  box-shadow: 0 6px 18px rgba(37, 99, 235, 0.5);
+  transform: translateY(-1px);
+}
+
+.status {
+  font-size: 0.9rem;
+}
+
+.status.ok {
+  color: #16a34a;
+}
+
+.status.error {
+  color: #dc2626;
+}
+
+.sidebar-section {
+  border-radius: 10px;
+  border: 1px solid #e5e7eb;
+  padding: 10px 12px;
+  background: #f9fafb;
+}
+
+.sidebar-section h2 {
+  font-size: 1rem;
+  margin: 0 0 6px;
+}
+
+.sidebar-row {
+  display: flex;
+  flex-direction: column;
+  gap: 4px;
+  margin-top: 6px;
+}
+
+.sidebar-row input[type="text"] {
+  padding: 6px 8px;
+  border-radius: 6px;
+  border: 1px solid #d1d5db;
+  font-size: 0.9rem;
+}
+
+.detail-value {
+  padding: 4px 6px;
+  border-radius: 6px;
+  border: 1px solid #e5e7eb;
+  background: #f9fafb;
+  font-size: 0.9rem;
+  min-height: 24px;
+}
+
+.objects-list {
+  display: flex;
+  flex-direction: column;
+  gap: 8px;
+  max-height: 260px;
+  overflow-y: auto;
+  padding-right: 4px;
+}
+
+.objects-tags {
+  margin-top: 8px;
+  display: flex;
+  flex-wrap: wrap;
+  gap: 4px;
+}
+
+.object-tag {
+  padding: 2px 6px;
+  border-radius: 999px;
+  background: #e5e7eb;
+  font-size: 0.75rem;
+  white-space: nowrap;
+}
+
+.object-item {
+  display: flex;
+  flex-direction: column;
+  gap: 4px;
+  padding: 4px 6px;
+  border-radius: 8px;
+  border: 1px solid #e5e7eb;
+  background: #ffffff;
+}
+
+.object-item-header {
+  display: flex;
+  align-items: center;
+  gap: 4px;
+  flex-wrap: nowrap;
+}
+
+.object-hierarchy-select {
+  width: 40px;
+  min-width: 0;
+  padding: 2px 3px;
+  border-radius: 6px;
+  border: 1px solid #d1d5db;
+  font-size: 0.8rem;
+  background: #f9fafb;
+}
+
+.object-parent-select {
+  width: 50px;
+  min-width: 0;
+  padding: 2px 3px;
+  border-radius: 6px;
+  border: 1px solid #d1d5db;
+  font-size: 0.8rem;
+  background: #eef2ff;
+}
+
+.object-item img {
+  width: 40px;
+  height: 40px;
+  object-fit: cover;
+  border-radius: 6px;
+  border: 1px solid #e5e7eb;
+}
+
+.object-item-text {
+  display: flex;
+  flex-direction: column;
+  font-size: 0.8rem;
+}
+
+.object-item-details {
+  padding-left: 24px;
+  display: none;
+  flex-direction: column;
+  gap: 4px;
+  margin-top: 4px;
+}
+
+.object-item-details.visible {
+  display: flex;
+}
+
+.object-item-details input[type="text"] {
+  padding: 4px 6px;
+  border-radius: 6px;
+  border: 1px solid #d1d5db;
+  font-size: 0.8rem;
+}
+
+.object-item-details label {
+  font-size: 0.75rem;
+}
+
+.sentences-pane {
+  flex: 0 0 300px;
+  display: flex;
+  flex-direction: column;
+  gap: 12px;
+}
+
+.sentences-list {
+  display: flex;
+  flex-direction: column;
+  gap: 12px;
+  max-height: 70vh;
+  overflow-y: auto;
+  padding: 4px;
+}
+
+.sentence-item {
+  padding: 12px;
+  background: #f9fbff;
+  border: 1px solid #d4ddf5;
+  border-radius: 8px;
+  display: flex;
+  flex-direction: column;
+  gap: 8px;
+}
+
+.sentence-item-question {
+  font-weight: 600;
+  color: #1e40af;
+  font-size: 0.95rem;
+}
+
+.sentence-item-answer {
+  color: #475569;
+  font-size: 0.9rem;
+  padding-left: 12px;
+  border-left: 2px solid #cbd5f0;
+}
+
+.sentence-item-empty {
+  padding: 24px;
+  text-align: center;
+  color: #94a3b8;
+  font-style: italic;
+}
+
+.object-item-text strong {
+  font-size: 0.85rem;
+}
+
+.selections-list {
+  display: flex;
+  flex-direction: column;
+  gap: 8px;
+  max-height: 200px;
+  overflow-y: auto;
+  padding: 8px;
+  background: #f9fbff;
+  border: 1px solid #d4ddf5;
+  border-radius: 8px;
+  margin-bottom: 8px;
+}
+
+.selection-item {
+  padding: 8px;
+  background: #ffffff;
+  border: 1px solid #e5e7eb;
+  border-radius: 6px;
+  font-size: 0.85rem;
+  color: #475569;
+}
+
+.selection-item strong {
+  color: #1e40af;
+  font-weight: 600;
+}
+
+.selections-empty {
+  padding: 16px;
+  text-align: center;
+  color: #94a3b8;
+  font-style: italic;
+  font-size: 0.85rem;
+}
--- a/templates/generate.html
+++ b/templates/generate.html
@@ -0,0 +1,159 @@
+<!DOCTYPE html>
+<html lang="de">
+  <head>
+    <meta charset="UTF-8" />
+    <title>GenerateIt</title>
+    <meta name="viewport" content="width=device-width, initial-scale=1" />
+    <link rel="stylesheet" href="/static/style.css" />
+  </head>
+  <body>
+    <div class="container">
+      <h1>GenerateIt</h1>
+
+      <div class="panel image-nav">
+        <div class="image-nav-left">
+          <button id="prevImageBtn" type="button">⬅️</button>
+          <span>Bild: <code id="currentImageName">{{ current_image or "–" }}</code></span>
+          <button id="nextImageBtn" type="button">➡️</button>
+          <button id="generateDetailsBtn" type="button">✨ KI‑Details</button>
+          <button id="generateSentenceBtn" type="button">💬 KI‑Sentence</button>
+        </div>
+        <div class="page-switch">
+          <select id="pageSelect">
+            <option value="/draw">DrawIt</option>
+            <option value="/generate" selected>GenerateIt</option>
+          </select>
+        </div>
+      </div>
+
+      <div class="main-layout">
+        <div class="objects-pane sidebar-section">
+          <h2>Objekte zu diesem Bild</h2>
+          <div id="objectsList" class="objects-list">
+            <!-- Wird per JS gefüllt -->
+          </div>
+          <div id="objectsTags" class="objects-tags">
+            <!-- Tags mit Objektnamen -->
+          </div>
+        </div>
+
+        <div class="right-pane">
+          <div class="sidebar-section">
+            <h2>Details</h2>
+            <div class="sidebar-row">
+              <label>Titel</label>
+              <div id="detailTitle" class="detail-value"></div>
+            </div>
+            <div class="sidebar-row">
+              <label>Position</label>
+              <div id="detailPosition" class="detail-value"></div>
+            </div>
+            <div class="sidebar-row">
+              <label>Status (sitzt/schwimmt/segelt)</label>
+              <div id="detailAction" class="detail-value"></div>
+            </div>
+            <div class="sidebar-row">
+              <label>Zustand (alt/jung/rostig)</label>
+              <div id="detailCondition" class="detail-value"></div>
+            </div>
+            <div class="sidebar-row">
+              <label>Hierarchie</label>
+              <div id="detailHierarchy" class="detail-value"></div>
+            </div>
+            <div class="sidebar-row">
+              <label>Gehört zu (Parent-Index)</label>
+              <div id="detailParent" class="detail-value"></div>
+            </div>
+          </div>
+
+          <div class="sidebar-section">
+            <h2>KI‑Details</h2>
+            <div class="sidebar-row">
+              <label>Label (EN)</label>
+              <div id="detailLabelEn" class="detail-value"></div>
+            </div>
+            <div class="sidebar-row">
+              <label>Label (DE)</label>
+              <div id="detailLabelDe" class="detail-value"></div>
+            </div>
+            <div class="sidebar-row">
+              <label>Label (SE)</label>
+              <div id="detailLabelSe" class="detail-value"></div>
+            </div>
+            <div class="sidebar-row">
+              <label>Farbe (EN)</label>
+              <div id="detailColorEn" class="detail-value"></div>
+            </div>
+            <div class="sidebar-row">
+              <label>Adjektiv (EN)</label>
+              <div id="detailAdjectiveEn" class="detail-value"></div>
+            </div>
+            <div class="sidebar-row">
+              <label>Action Verb (EN)</label>
+              <div id="detailActionVerbEn" class="detail-value"></div>
+            </div>
+            <div class="sidebar-row">
+              <label>Präposition (EN)</label>
+              <div id="detailPrepositionEn" class="detail-value"></div>
+            </div>
+            <div class="sidebar-row">
+              <label>Relative Position (EN)</label>
+              <div id="detailRelativePositionEn" class="detail-value"></div>
+            </div>
+            <div class="sidebar-row">
+              <label>Season (EN)</label>
+              <div id="detailSeasonEn" class="detail-value"></div>
+            </div>
+          </div>
+
+          <div class="sidebar-section">
+            <h2>KI‑Sentence</h2>
+            <h3 style="font-size: 0.9rem; margin-top: 0; margin-bottom: 8px; color: #475569;">Einfach (für Kinder)</h3>
+            <div class="sidebar-row">
+              <label>Question (EN)</label>
+              <div id="detailSentenceQuestionSimple" class="detail-value"></div>
+            </div>
+            <div class="sidebar-row">
+              <label>Answer (EN)</label>
+              <div id="detailSentenceAnswerSimple" class="detail-value"></div>
+            </div>
+            <h3 style="font-size: 0.9rem; margin-top: 12px; margin-bottom: 8px; color: #475569;">Fortgeschritten</h3>
+            <div class="sidebar-row">
+              <label>Question (EN)</label>
+              <div id="detailSentenceQuestionAdvanced" class="detail-value"></div>
+            </div>
+            <div class="sidebar-row">
+              <label>Answer (EN)</label>
+              <div id="detailSentenceAnswerAdvanced" class="detail-value"></div>
+            </div>
+          </div>
+        </div>
+
+        <div class="sentences-pane">
+          <div class="sidebar-section">
+            <h2>Alle Sätze</h2>
+            <div id="sentencesList" class="sentences-list">
+              <!-- Wird per JS gefüllt -->
+            </div>
+          </div>
+        </div>
+      </div>
+    </div>
+
+    <script>
+      (function () {
+        window.isGeneratePage = true;
+        const pageSelect = document.getElementById("pageSelect");
+        if (pageSelect) {
+          pageSelect.addEventListener("change", (e) => {
+            window.location.href = e.target.value;
+          });
+        }
+        window.initialImages = {{ images|tojson|safe }};
+        window.initialImageIndex = {{ current_index if current_index is not none else -1 }};
+      })();
+    </script>
+    <script src="/static/script.js"></script>
+  </body>
+</html>
+
--- a/templates/index.html
+++ b/templates/index.html
@@ -0,0 +1,124 @@
+<!DOCTYPE html>
+<html lang="de">
+  <head>
+    <meta charset="UTF-8" />
+    <title>Bild-Ausschnitt wählen</title>
+    <meta name="viewport" content="width=device-width, initial-scale=1" />
+    <link rel="stylesheet" href="/static/style.css" />
+  </head>
+  <body>
+    <div class="container">
+      <h1>Bild-Ausschnitt wählen</h1>
+
+      <div class="panel image-nav">
+        <div class="image-nav-left">
+          <button id="prevImageBtn" type="button">⬅️</button>
+          <span>Bild: <code id="currentImageName">{{ current_image or "–" }}</code></span>
+          <button id="nextImageBtn" type="button">➡️</button>
+          <button id="saveImageBtn" type="button">💾</button>
+        </div>
+        <div class="page-switch">
+          <select id="pageSelect">
+            <option value="/draw" selected>DrawIt</option>
+            <option value="/generate">GenerateIt</option>
+          </select>
+        </div>
+      </div>
+
+      <div class="main-layout">
+        <div class="objects-pane sidebar-section">
+          <h2>Objekte zu diesem Bild</h2>
+          <div id="objectsList" class="objects-list">
+            <!-- Wird per JS gefüllt -->
+          </div>
+          <div id="objectsTags" class="objects-tags">
+            <!-- Tags mit Objektnamen -->
+          </div>
+        </div>
+
+        <div class="left-pane">
+          <div class="canvas-wrapper">
+            <canvas id="imageCanvas"></canvas>
+          </div>
+        </div>
+
+        <div class="right-pane">
+          <div class="sidebar-section">
+            <h2>Auswahl</h2>
+            <div class="sidebar-row">
+              <span>Auswahl-Typ (Interface / Backend):</span>
+            </div>
+            <div class="sidebar-row">
+              <label class="mode-option">
+                <input type="radio" name="mode" value="rect" checked />
+                Rechteck / <code>BBox</code>
+              </label>
+            </div>
+            <div class="sidebar-row">
+              <label class="mode-option">
+                <input type="radio" name="mode" value="polygon" />
+                Polygon / <code>polygon</code>
+              </label>
+            </div>
+            <div class="sidebar-row">
+              <button id="clearSelectionBtn" type="button">Auswahl zurücksetzen</button>
+            </div>
+          </div>
+
+          <div class="sidebar-section">
+            <h2>Metadaten</h2>
+            <div class="sidebar-row">
+              <label for="title_de">Titel / <code>title_de</code></label>
+              <input id="title_de" type="text" />
+            </div>
+            <div class="sidebar-row">
+              <label for="position_de">Position / <code>position_de</code></label>
+              <input id="position_de" type="text" />
+            </div>
+            <div class="sidebar-row">
+              <label for="action_de">Status (sitzt/schwimmt/segelt) / <code>action_de</code></label>
+              <input id="action_de" type="text" placeholder="z.B. sitzt" />
+            </div>
+            <div class="sidebar-row">
+              <label for="condition_de">Zustand (alt/jung/rostig) / <code>condition_de</code></label>
+              <input id="condition_de" type="text" placeholder="z.B. rostig" />
+            </div>
+          </div>
+
+          <div class="sidebar-section">
+            <h2>Auswahlen</h2>
+            <div id="selectionsList" class="selections-list">
+              <div class="selections-empty">Noch keine Auswahlen hinzugefügt.</div>
+            </div>
+            <div class="sidebar-row">
+              <button id="addSelectionBtn" disabled>➕ Auswahl hinzufügen</button>
+            </div>
+            <div class="sidebar-row">
+              <button id="saveCropBtn" disabled>💾 Objekt speichern</button>
+            </div>
+            <div class="sidebar-row">
+              <button id="clearAllSelectionsBtn" type="button">🗑️ Alle Auswahlen löschen</button>
+            </div>
+            <span id="status"></span>
+          </div>
+        </div>
+      </div>
+    </div>
+
+    <script>
+      (function () {
+        window.isGeneratePage = false;
+        const pageSelect = document.getElementById("pageSelect");
+        if (pageSelect) {
+          pageSelect.addEventListener("change", (e) => {
+            window.location.href = e.target.value;
+          });
+        }
+        window.initialImages = {{ images|tojson|safe }};
+        window.initialImageIndex = {{ current_index if current_index is not none else -1 }};
+      })();
+    </script>
+    <script src="/static/script.js"></script>
+  </body>
+</html>
+