ComfyUI Node
Open-H3 Regenerate 2K
A ComfyUI node in 🧪AILab/Open-H3 with 8 inputs and 3 outputs.
Open-H3 Regenerate 2K
- model
- latent
- MODEL
- LATENT
- SIGMAS
â—„model_nameOpen-H3-Regenerate-2K (BF16)â–º
â—„strength1.00â–º
â—„sharp_curvetrueâ–º
â—„steps3â–º
â—„lock_audiotrueâ–º
â—„unload_modelfalseâ–º
Category🧪AILab/Open-H3
Inputs (8)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | MiniMax H3 base model (minimax_h3_ref2va_pruned_int8_convrot.safetensors). | |
| model_name | COMBO | Open-H3-Regenerate-2K (BF16) | Select Open-H3 Regenerate 2K model preset. Auto-downloads from 1038lab/Open-H3-Regenerate-2K to ComfyUI/models/loras/minmax/ if missing. |
| strength | FLOAT | 1.000–1.5 | Patch strength applied to the Open-H3 2K model and Sharp AdaLN curve branches. |
| sharp_curve | BOOLEAN | true | Load the 50 LMS Sharp AdaLN curve branches (sharp_curve_branches.safetensors) for enhanced 2K sharpness. |
| steps | COMBO | 3 | Official calibrated Euler refinement schedule: 3 steps for best stability, 2 steps for speed. |
| lock_audio | BOOLEAN | true | Lock the audio stream in the joint H3 latent at sigma 0 so only video frames are regenerated. |
| unload_model | BOOLEAN | false | Offload cached models and flush GPU VRAM immediately after patching. |
| latentopt | LATENT | Joint MiniMax H3 video/audio latent (av_latent) to refine at 2K resolution. |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| MODEL | MODEL | — |
| LATENT | LATENT | — |
| SIGMAS | SIGMAS | — |