ComfyUI Node
H3 • Models
Load one shared H3 model, Qwen encoder and VIDEO VAE for generation and editing. No image VAE, hybrid model, audio VAE or refiner is required.
H3 • Models
- models
◄diffusion_model▾►
◄text_encoder▾►
◄video_vae▾►
◄turbo_loraNone►
CategoryMiniMax H3/Image Studio
Inputs (4)
| Name | Type | Default | Description |
|---|---|---|---|
| diffusion_model | COMBO | Official FL2VA is the single starting model for both workflows, following Fizgig. REF2VA can also be selected for edit comparisons. | |
| text_encoder | COMBO | MiniMax H3 Qwen text/vision encoder; shared by both workflows. | |
| video_vae | COMBO | Official H3 video VAE, INT8 ConvRot or FP16. The still decoder is included. | |
| turbo_loraopt | COMBO | None | Compatibility shortcut for Larry v4 at 0.38. Prefer chained H3 LoRA nodes for manual strengths/multiple adapters. Leave None when applying Larry through a LoRA node; do not load it twice. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| models | H3_STUDIO_MODELS | Shared H3 models. Connect to Text to Image or Image Edit. |