ComfyUI Node
IronLeo78 MTV Omni Model Loader
Model Loader for the IronLeo78 MTV Omni. Plug 'models' into the MTV Omni's optional 'models' input to change the diffusion model, text encoder, video / audio VAE, three optional custom LoRAs LoRa_1-3 (toggle, file, strength each), the step count and the noise seed. Nothing connected = the MTV Omni uses its hardcoded models (FastH3 OmniRef, 8 steps). by IronLeo78
IronLeo78 MTV Omni Model Loader
- models
◄unet_nameNone►
◄qwen_vl_nameNone►
◄video_vae_namepixel_space►
◄audio_vae_namepixel_space►
◄LoRa_1_enabledfalse►
◄LoRa_1None►
◄LoRa_1_strength1.00►
◄LoRa_2_enabledfalse►
◄LoRa_2None►
◄LoRa_2_strength1.00►
◄LoRa_3_enabledfalse►
◄LoRa_3None►
◄LoRa_3_strength1.00►
◄steps8►
◄noise_seed1978►
CategoryIronLeo78/MTV Omni
Inputs (15)
| Name | Type | Default | Description |
|---|---|---|---|
| unet_name | COMBO | None | Diffusion model. Files from ComfyUI/models/diffusion_models/. The MTV Omni's hardcoded file is fasth3_omniref_8step_pruned_int8_convrot.safetensors. |
| qwen_vl_name | COMBO | None | Qwen3-VL text encoder. Files from ComfyUI/models/text_encoders/. The MTV Omni's hardcoded file is qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors. |
| video_vae_name | COMBO | pixel_space | Video VAE. Files from ComfyUI/models/vae/. The MTV Omni's hardcoded file is minimax_h3_video_vae_int8_convrot.safetensors. |
| audio_vae_name | COMBO | pixel_space | Audio VAE. Files from ComfyUI/models/vae/. The MTV Omni's hardcoded file is minimax_h3_audio_vae_fp32.safetensors. |
| LoRa_1_enabled | BOOLEAN | false | LoRa_1 ON / OFF (default OFF). Applied right after the diffusion model. A LoRA trained for the base H3 model is an experiment on the distilled OmniRef student. |
| LoRa_1 | COMBO | None | LoRa_1: LoRA file from ComfyUI/models/loras/. |
| LoRa_1_strength | FLOAT | 1.00-100–100 | LoRa_1 strength (default 1.0). |
| LoRa_2_enabled | BOOLEAN | false | LoRa_2 ON / OFF (default OFF). Applied right after the diffusion model and after LoRa_1. A LoRA trained for the base H3 model is an experiment on the distilled OmniRef student. |
| LoRa_2 | COMBO | None | LoRa_2: LoRA file from ComfyUI/models/loras/. |
| LoRa_2_strength | FLOAT | 1.00-100–100 | LoRa_2 strength (default 1.0). |
| LoRa_3_enabled | BOOLEAN | false | LoRa_3 ON / OFF (default OFF). Applied right after the diffusion model and after LoRa_2. A LoRA trained for the base H3 model is an experiment on the distilled OmniRef student. |
| LoRa_3 | COMBO | None | LoRa_3: LoRA file from ComfyUI/models/loras/. |
| LoRa_3_strength | FLOAT | 1.00-100–100 | LoRa_3 strength (default 1.0). |
| steps | INT | 81–200 | Sampling steps (hardcoded default 8: FastH3 OmniRef is distilled for 8 steps; other values are an experiment). |
| noise_seed | INT | 19780–18446744073709550000 | RandomNoise seed of the MTV Omni (default 1978, control after generate = fixed). Used for the sampling pass. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| models | IRONLEO78_MTV_MODELS | — |