Pick a checkpoint
H3-Base-FL2VA covers text-to-video and first/last-frame inputs. H3-Base-Ref2VA covers multi-reference image, video, and audio inputs.
Serve the base model
The official model card documents SGLang, vLLM, Diffusers, and ComfyUI paths. The SGLang examples use separate FL2VA and Ref2VA services.
Know the boundary
H3-Context-IR and H3-Regenerate-2K are not included in the open release. Local deployment validates 768P; the full official 2K workflow still uses hosted API modules.
Review before shipping
Check the current Community License, GPU memory, checkpoint storage, private-data boundary, and whether an API call is acceptable for your deployment.