- mcp_server: add explicit required/optional markers to all tool params, improve reroll schema to auto-convert required to optional-override, update default neg prompt
- history: fix next_id to scan global max across all records, add file fallback search when record not found in memory
Model configuration:
- Add COMFYUI_MODELS_DIR for model pre-check (optional)
- Add ANIMATOOL_UNET_NAME, ANIMATOOL_CLIP_NAME, ANIMATOOL_VAE_NAME
- Add ANIMATOOL_CHECK_MODELS toggle
Model pre-check:
- Check if model files exist before generation
- Friendly error message with download link
- Skipped when COMFYUI_MODELS_DIR not set (remote ComfyUI)
Other improvements:
- Use config model names as defaults in _inject()
- Export DEFAULT_*_NAME constants
Anima model (based on Cosmos architecture) requires latent space dimensions to be divisible by spatial_patch_size=2. Since VAE scales down by 8x, pixel dimensions must be divisible by 8*2=16.
Changes:
- Update round_to default from 8 to 16 in config.py and anima_executor.py
- Add align_dimension() function to align user-provided width/height
- Apply alignment to directly specified width/height (not just aspect_ratio)
- Update schema descriptions to document 16-multiple requirement
- MCP Server: Native image display in Cursor/Claude
- ComfyUI HTTP API: /anima/health, /anima/schema, /anima/knowledge, /anima/generate
- Standalone FastAPI server and CLI tool
- Structured prompt system with 14 aspect ratio presets
- Expert knowledge base for Anima prompting
Nakano Kenji