image-generationlisted

# Image Generation (ComfyUI + FLUX) ## Folder structure The portable Windows release unpacks as: ``` ComfyUI/ ← outer folder (gitignored) ├── run_nvidia_gpu.bat ├── python_embeded/ └── ComfyUI/ ← inner folder — the actual app └── models/ ├── checkpoints/ ← flux1-schnell-fp8.safetensors (full checkpoint from Comfy-Org) ├── vae/ ← ae.safetensors └── clip/ ← clip_l.safetensors, t5xxl_fp8_e4m3fn.safetensors ``` Models must go in `ComfyUI/ComfyUI/models/`, not `ComfyUI/models/`. If ComfyUI returns 400 with empty node lists, the models are in the wrong folder. ## Scripts - `scripts/utils/generate_image.py` — single image, CLI args: `outputPath WxH prompt` - `scripts/utils/generate_images_batch.py` — reads `.tmp/images.json`, calls the above per entry Run: `.venv\Scripts\python.exe scripts\utils\generate_images_batch.py .tmp\images.json` ComfyUI must be running at `http://127.0.0.1:8188` before invoking either script. ## Windows Unicode encoding Prompts that contain non-ASCII characters (e.g. accented names from localization) crash the batch script on Windows due to the cp1251 console codepage. Always prefix the run command with `$env:PYTHONUTF8 = '1'`: ```powershell $env:PYTHONUTF8 = '1'; & ".venv\Scripts\python.exe" "scripts\utils\generate_images_batch.py" ".tmp\images.json" ``` ## Character portrait recipe - **Output path:** `Assets/Textures/Characters/{characterId}.png` - **Size:**