Spaces:
Sleeping
Sleeping
Commit History
9 days ago
-
Load Comfy-Org's pruned int8_convrot diffusion model (H3_DIFFUSION_MODEL switch)
07dc2bf -
Read the preloaded models from the inherited global, not a pickled argument
fc014ea -
Preload models outside the GPU window, recalibrate the budget, Lanczos keyframes, gate SageAttention
89298ca
-
Strip to the I2VA/FLF2VA path with a single generate_video API
Drop T2VA, REF2VA, keyframe guides, ControlNet, LoRA chains, the MCP tool server and the YAML recipe/injector framework. One app.py builds the MiniMaxH3ImageToVideo graph directly, declares a per-job ZeroGPU budget (mirrored in protocol-generator), and persists finished videos to the igitagoz/h3-job-outputs dataset keyed by job_id. ComfyUI is pinned to c75d8c9 and requirements.txt carries that commit's own pins. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>a2c9f20
-
feat: upgrade to Gradio 6 and merge tabs into unified multi-task UI
Verified signature439fb79
-
Initial project: MiniMax-H3
Verified signature6ac5e98