Add FLUX.2 Klein 4B image generation tool with scene presets (v0.11.0)

- New tools/flux2.py: text-to-image, image editing, multi-image compositing
- 8 scene presets for video production (title-bg, problem, solution, etc.)
- Brand-aware generation reads brand.json colors into prompts
- Docker image with baked model weights (ghcr.io/conalmullan/video-toolkit-flux2)
- Rebuilt qwen3-tts Docker image with baked models
- "The Space Between" showcase video (all AI-generated)
- Updated README, registry, and changelog

Closes #6

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: ConalMullan

README.md

@@ -22,6 +22,7 @@ An AI-native video production workspace for [Claude Code](https://claude.ai/code
 > - `ghcr.io/conalmullan/video-toolkit-propainter` — Video watermark removal (ProPainter)
 > - `ghcr.io/conalmullan/video-toolkit-sadtalker` — Talking head generation (SadTalker)
 > - `ghcr.io/conalmullan/video-toolkit-qwen3-tts` — Text-to-speech (Qwen3-TTS)
+> - `ghcr.io/conalmullan/video-toolkit-flux2` — Text-to-image & editing (FLUX.2 Klein 4B)
 >
 > My motto: **Be brave. Experiment.** And please share any videos you create or ideas you have back with the project — it helps me keep improving this toolkit for everyone.
 
@@ -149,6 +150,7 @@ See `examples/` for finished projects you can learn from (oldest first, showing
 | 2026-01-22 | [ds-remote-mcp](https://demos.digitalsamba.com/video/ds-remote-mcp.mp4) | Remote MCP server demo *(the jazz background music is a joke)* |
 | 2026-01-25 | [schlumbergera](https://demos.digitalsamba.com/video/schlumbergera.mp4) | Android sprint review video |
 | 2026-02-23 | [cortina](https://demos.digitalsamba.com/video/sprint-review.mp4) | Mobile platforms sprint review |
+| 2026-03-15 | [the-space-between](https://demos.digitalsamba.com/video/the-space-between.mp4) | AI-generated video essay — flux2 avatar, Qwen3-TTS voice, SadTalker animation |
 
 ### Scene Transitions
 
@@ -250,6 +252,12 @@ python tools/locate_watermark.py --input video.mp4 --grid --output-dir ./review/
 # Creates animated presenter/narrator from a static portrait + voiceover audio
 # Use with NarratorPiP component for picture-in-picture presenter overlays
 python tools/sadtalker.py --image portrait.png --audio voiceover.mp3 --output talking.mp4
+
+# AI image generation (FLUX.2 Klein 4B — text-to-image + editing)
+python tools/flux2.py --prompt "A sunset over mountains"
+python tools/flux2.py --preset title-bg --brand digital-samba
+python tools/flux2.py --input photo.jpg --prompt "Add sunglasses"
+python tools/flux2.py --list-presets
 ```
 
 **Tool Categories:**
@@ -258,7 +266,7 @@ python tools/sadtalker.py --image portrait.png --audio voiceover.mp3 --output ta
 |------|-------|---------|
 | **Project** | voiceover, music, sfx | Used during video creation workflow |
 | **Utility** | redub, addmusic, notebooklm_brand, locate_watermark | Quick transformations, no project needed |
-| **Cloud GPU** | image_edit, upscale, dewatermark, sadtalker, qwen3_tts | AI processing via RunPod (see below) |
+| **Cloud GPU** | image_edit, upscale, dewatermark, sadtalker, qwen3_tts, flux2 | AI processing via RunPod (see below) |
 
 See [docs/runpod-setup.md](docs/runpod-setup.md) for Cloud GPU tool setup.
 
@@ -273,6 +281,7 @@ Cloud GPU tools use pre-built Docker images deployed to RunPod serverless:
 | dewatermark | `ghcr.io/conalmullan/video-toolkit-propainter:latest` | 24GB (RTX 3090/4090) |
 | sadtalker | `ghcr.io/conalmullan/video-toolkit-sadtalker:latest` | 24GB (RTX 4090) |
 | qwen3_tts | `ghcr.io/conalmullan/video-toolkit-qwen3-tts:latest` | 24GB (ADA) |
+| flux2 | `ghcr.io/conalmullan/video-toolkit-flux2:latest` | 24GB (ADA) |
 
 Dockerfiles and handlers are in `docker/`. Run `python tools/<tool>.py --setup` to auto-deploy.
 

_internal/CHANGELOG.md

@@ -6,6 +6,32 @@ All notable changes to claude-code-video-toolkit.
 
 ---
 
+## 2026-03-15 (v0.11.0)
+
+### Added
+- **`tools/flux2.py`** — AI image generation and editing using FLUX.2 Klein 4B
+  - Text-to-image generation from prompts (~2.5s fast mode, ~8s quality mode)
+  - Image editing with reference images (`--input photo.jpg --prompt "..."`)
+  - Multi-image compositing (up to 3 reference images)
+  - **8 scene presets** for video production: `title-bg`, `problem`, `solution`, `demo-bg`, `stats-bg`, `cta`, `thumbnail`, `portrait-bg`
+  - **Brand-aware generation** — `--brand digital-samba` reads brand.json and injects color palette into prompts
+  - Preset + prompt layering — preset provides style/mood, `--prompt` adds subject context
+  - `--setup` creates RunPod template + endpoint via GraphQL
+  - `--list-presets` shows all available scene presets
+  - Docker image: `ghcr.io/conalmullan/video-toolkit-flux2:latest` (baked model weights, ~15GB)
+  - Apache 2.0 licensed model (commercial OK)
+- **"The Space Between"** — [showcase video](https://demos.digitalsamba.com/video/the-space-between.mp4) demonstrating end-to-end AI video creation
+  - Avatar generated with flux2, voiced with Qwen3-TTS, animated with SadTalker, composed in Remotion
+  - Concept, script, voice, visuals — all AI-generated
+  - New video format: video essay (beyond sprint reviews and product demos)
+
+### Changed
+- **Baked Docker images rebuilt** — both `flux2` and `qwen3-tts` images now include model weights baked in, eliminating cold-start model downloads (~30s cold start vs minutes)
+- Updated `_internal/toolkit-registry.json` with flux2 tool entry, cloud endpoint, scene presets, and showcase example
+- Updated README with flux2 tool, Docker image, demo table entry, and Cloud GPU tools table
+
+---
+
 ## 2026-02-25 (v0.10.1)
 
 ### Added

_internal/toolkit-registry.json

@@ -1,6 +1,6 @@
 {
   "name": "claude-code-video-toolkit",
-  "version": "0.10.1",
+  "version": "0.11.0",
   "description": "AI-native video production workspace for Claude Code",
   "repository": "https://github.com/digitalsamba/claude-code-video-toolkit",
 
@@ -309,6 +309,30 @@
       "requires": "ffprobe (from ffmpeg)",
       "created": "2026-02-24",
       "updated": "2026-02-24"
+    },
+    "flux2": {
+      "path": "tools/flux2.py",
+      "description": "AI image generation and editing using FLUX.2 Klein 4B - text-to-image, image editing, and scene presets",
+      "usage": "python tools/flux2.py --preset title-bg --brand digital-samba",
+      "status": "beta",
+      "category": "image-generation",
+      "backend": "flux2-klein-4b",
+      "requires": "RunPod account",
+      "options": {
+        "modes": ["generate", "edit"],
+        "presets": ["title-bg", "problem", "solution", "demo-bg", "stats-bg", "cta", "thumbnail", "portrait-bg"],
+        "brand": true,
+        "width": [512, 768, 1024, 1280, 1920],
+        "height": [512, 768, 1024, 1080, 1280],
+        "steps": "4 (fast) to 50 (quality)",
+        "guidance": "1.0 (fast) to 4.0 (quality)",
+        "seed": true,
+        "multiImage": true
+      },
+      "envVars": ["RUNPOD_API_KEY", "RUNPOD_FLUX2_ENDPOINT_ID"],
+      "estimatedCost": "$0.01-0.03 per image",
+      "created": "2026-03-14",
+      "updated": "2026-03-15"
     }
   },
 
@@ -371,10 +395,17 @@
           "envVar": "RUNPOD_QWEN3_TTS_ENDPOINT_ID",
           "operations": ["qwen3_tts"],
           "estimatedCost": "$0.005-0.10 per generation"
+        },
+        "flux2": {
+          "image": "ghcr.io/conalmullan/video-toolkit-flux2:latest",
+          "dockerfile": "docker/runpod-flux2/",
+          "envVar": "RUNPOD_FLUX2_ENDPOINT_ID",
+          "operations": ["flux2"],
+          "estimatedCost": "$0.01-0.03 per image"
         }
       },
       "created": "2025-12-30",
-      "updated": "2026-01-12"
+      "updated": "2026-03-14"
     }
   },
 
@@ -561,6 +592,14 @@
       "template": "sprint-review",
       "complexity": "intermediate",
       "contributor": "Digital Samba"
+    },
+    "the-space-between": {
+      "path": "examples/the-space-between/",
+      "description": "AI-generated video essay — flux2 avatar, Qwen3-TTS voice, SadTalker animation",
+      "template": "custom",
+      "complexity": "advanced",
+      "demo": "https://demos.digitalsamba.com/video/the-space-between.mp4",
+      "created": "2026-03-15"
     }
   },
 

docker/runpod-flux2/Dockerfile

@@ -0,0 +1,91 @@
+# RunPod Serverless handler for FLUX.2 Klein 4B (models baked in)
+#
+# Build: docker buildx build --platform linux/amd64 -t ghcr.io/conalmullan/video-toolkit-flux2:latest --push .
+#
+# Image size: ~15GB (includes ~8GB model weights)
+# Cold start: ~30s (no model download needed)
+#
+# GPU Requirements:
+#   - Minimum: 16GB VRAM (RTX 4070 Ti, A4000)
+#   - Optimal: 24GB VRAM (RTX 4090, ADA_24)
+
+FROM nvidia/cuda:12.4.1-cudnn-devel-ubuntu22.04
+
+ENV DEBIAN_FRONTEND=noninteractive
+
+# Install Python and system dependencies
+RUN apt-get update && apt-get install -y \
+    python3.11 \
+    python3-pip \
+    python3.11-venv \
+    python3.11-dev \
+    git \
+    git-lfs \
+    curl \
+    && rm -rf /var/lib/apt/lists/* \
+    && ln -sf /usr/bin/python3.11 /usr/bin/python3 \
+    && ln -sf /usr/bin/python3.11 /usr/bin/python
+
+RUN git lfs install
+
+WORKDIR /app
+
+# CUDA environment
+ENV CUDA_HOME=/usr/local/cuda
+ENV PATH="${CUDA_HOME}/bin:${PATH}"
+ENV LD_LIBRARY_PATH="${CUDA_HOME}/lib64:${LD_LIBRARY_PATH}"
+
+# Install PyTorch with CUDA 12.4 support
+RUN pip3 install --no-cache-dir \
+    torch==2.5.1 \
+    torchvision==0.20.1 \
+    --index-url https://download.pytorch.org/whl/cu124
+
+# Install diffusers from git (required for Flux2KleinPipeline)
+RUN pip3 install --no-cache-dir git+https://github.com/huggingface/diffusers
+
+# Install dependencies
+RUN pip3 install --no-cache-dir \
+    transformers>=4.45.0 \
+    accelerate>=0.30.0 \
+    safetensors>=0.4.0 \
+    sentencepiece>=0.2.0 \
+    protobuf>=4.25.0 \
+    Pillow>=10.0.0 \
+    numpy>=1.26.0
+
+# Install RunPod SDK and utilities
+RUN pip3 install --no-cache-dir \
+    runpod>=1.7.0 \
+    requests>=2.31.0 \
+    boto3>=1.34.0 \
+    huggingface_hub>=0.25.0
+
+# Set HF cache location
+ENV HF_HOME=/root/.cache/huggingface
+
+# === BAKE MODEL INTO IMAGE ===
+# Download FLUX.2 Klein 4B during build (~8GB)
+RUN python3 -c "\
+from huggingface_hub import snapshot_download; \
+import os; \
+os.makedirs('/root/.cache/huggingface', exist_ok=True); \
+print('Downloading FLUX.2 Klein 4B model...'); \
+snapshot_download('black-forest-labs/FLUX.2-klein-4B', cache_dir='/root/.cache/huggingface'); \
+print('Model downloaded successfully'); \
+"
+
+# Copy handler
+COPY handler.py /app/handler.py
+
+ENV PYTHONUNBUFFERED=1
+
+# Health check — verify torch and diffusers
+RUN python3 -c "\
+import torch; \
+from diffusers import Flux2KleinPipeline; \
+print(f'PyTorch {torch.__version__}, CUDA available: {torch.cuda.is_available()}'); \
+print('Flux2KleinPipeline: available'); \
+"
+
+CMD ["python3", "-u", "/app/handler.py"]