character-swap-h3
character-swap-h3
Run with python scripts/comfy_client.py --url $COMFY_URL run workflows/<file> --set NODE.INPUT=VALUE --out out/, and upload inputs first with comfy_client.py upload <file>. Before the first run, check the server with python scripts/preflight.py --url $COMFY_URL workflows/<file>. Node packs, model files, VRAM and weight licenses are listed in references/requirements.md. Use open-weights models only. LTX weights are under the Lightricks LTX license, which is not an OSI-approved open-source license.
About H3: these graphs load MiniMax H3 weights locally, from Comfy-Org's Hugging Face repos. They do not call a hosted MiniMax API, so this is an open-weights path. The H3 weight license is unverified, so check it before commercial use. If H3 weights aren't available, or their license doesn't fit, use the LTX fallback.
Install (once)
- Copy
assets/comfyui-h3-oneclick-swap/intoComfyUI/custom_nodes/. It is MIT, from D3voz. - The other packs (KJNodes, VideoHelperSuite, H3-FaceRefine, MiniMaxH3-Context-Loop, Fantastic PromptBuilder) are listed in
references/requirements.md. - H3-FaceRefine needs insightface. Its default
buffalo_lface models are commonly distributed for non-commercial research use (unverified), so check them before commercial work.
| Need | Workflow | Set |
|---|---|---|
| Swap a person (whole body) | h3_swap_simple.api.json |
5.video clip, 2.image reference, 11.seed, 11.resolution (736p) |
| Swap a face only | h3_swap_face.api.json |
1.video, 2.image, 24.noise_seed |
| Swap an object, outfit or product | h3_swap_object.api.json |
5.video, 2.image, 18.text (what to replace), 7.seed |
| Open fallback (LTX-2.3 masked motion swap) | ltx23_motion_swap_open_fallback.api.json |
1.file, 2.image, 3.text (who to replace), 5.noise_seed |