> ## Documentation Index
> Fetch the complete documentation index at: https://docs.comfy.org/llms.txt
> Use this file to discover all available pages before exploring further.

# ComfyUI MiniMax H3 Text-to-Video, Image-to-Video, and Reference-to-Video Workflows

> Step-by-step MiniMax H3 native workflow examples in ComfyUI, including guide anchoring and latent noise masks for finer control over video generation.

This page covers the three base MiniMax H3 templates plus advanced native-node techniques. For capabilities, tuning, and speedups, see the [MiniMax H3 overview](https://docs.comfy.org/tutorials/video/minimax/minimax-h3.md).

## MiniMax H3 Text to Video (T2V)

Generate videos from text prompts with native stereo audio.

<video controls src="https://raw.githubusercontent.com/Comfy-Org/workflow_templates/main/output/video_minimax_h3_t2v.mp4" />

> [card] [Run on Comfy Cloud](https://cloud.comfy.org/?template=video_minimax_h3_t2v\&utm_source=docs\&utm_medium=referral\&utm_campaign=minimax-h3)
>
> Open in Comfy Cloud

> [card] [Download Workflow](https://github.com/Comfy-Org/workflow_templates/blob/main/templates/video_minimax_h3_t2v.json)
>
> Download JSON or search "MiniMax H3 T2V" in Template Library

### Model downloads

> [card] [Diffusion Model: minimax\_h3\_fl2va\_pruned\_int8\_convrot](https://huggingface.co/Comfy-Org/MiniMax-H3/blob/main/diffusion_models/minimax_h3_fl2va_pruned_int8_convrot.safetensors)
>
> Place in `ComfyUI/models/diffusion_models/`

> [card] [Text Encoder: qwen3vl\_32b\_minimax\_h3\_nvfp4\_awq](https://huggingface.co/Comfy-Org/MiniMax-H3/blob/main/text_encoders/qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors)
>
> Place in `ComfyUI/models/text_encoders/`

> [card] [VAE: minimax\_h3\_video\_vae\_int8\_convrot](https://huggingface.co/Comfy-Org/MiniMax-H3/blob/main/vae/minimax_h3_video_vae_int8_convrot.safetensors)
>
> Place in `ComfyUI/models/vae/`

> [card] [VAE: minimax\_h3\_audio\_vae\_fp32](https://huggingface.co/Comfy-Org/MiniMax-H3/blob/main/vae/minimax_h3_audio_vae_fp32.safetensors)
>
> Place in `ComfyUI/models/vae/`

> [card] [LoRA: minimax\_h3\_fl2v\_turbo\_8step\_v1.0\_comfyui\_bf16](https://huggingface.co/lightx2v/Minimax-h3-Turbo/resolve/main/minimax_h3_fl2v_turbo_8step_v1.0_comfyui_bf16.safetensors)
>
> Required by the workflow's model scan; powers the optional Lightning LoRA turbo mode. Place in `ComfyUI/models/loras/`

### Model storage

```
ComfyUI/
├── 📂 models/
│   ├── 📂 diffusion_models/
│   │   └── minimax_h3_fl2va_pruned_int8_convrot.safetensors
│   ├── 📂 text_encoders/
│   │   └── qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors
│   ├── 📂 vae/
│   │   ├── minimax_h3_video_vae_int8_convrot.safetensors
│   │   └── minimax_h3_audio_vae_fp32.safetensors
│   ├── 📂 loras/
│   │   └── minimax_h3_fl2v_turbo_8step_v1.0_comfyui_bf16.safetensors
│   └── 📂 embeddings/
│       └── minimaxh3_art_is_explosion.safetensors
```

### Prompting tips

1. **Scene first**: Overall scene, then timed shots
2. **One block**: Shots, camera, dialogue, SFX, and music together
3. **Resolution**: Native 16:9 is **1344×768** (768px short edge); round to multiples of **32**
4. **Duration**: Snaps to **17k+5** frames at 24fps
5. **First/last frame**: Wire images to `first_frame` / `last_frame` on `MiniMaxH3ImageToVideo` for FL2VA
6. **Turbo**: Default **20** steps; enable `turbo_mode` for **8**-step Lightning LoRA (faster, slightly softer motion/audio). Raise steps (e.g. **25**) for motion quality

Official base guide: [prompt guide](https://docs.comfy.org/tutorials/video/minimax/minimax-h3-prompt-guide.md).

## MiniMax H3 Image to Video (I2V)

Generate videos from an input image, with optional first/last-frame keyframes.

<video controls src="https://raw.githubusercontent.com/Comfy-Org/workflow_templates/main/output/video_minimax_h3_i2v.mp4" />

> [card] [Run on Comfy Cloud](https://cloud.comfy.org/?template=video_minimax_h3_i2v\&utm_source=docs\&utm_medium=referral\&utm_campaign=minimax-h3)
>
> Open in Comfy Cloud

> [card] [Download Workflow](https://github.com/Comfy-Org/workflow_templates/blob/main/templates/video_minimax_h3_i2v.json)
>
> Download JSON or search "MiniMax H3 I2V" in Template Library

> [card] [Input Image: transparent\_rgb\_gaming\_mouse.png](https://raw.githubusercontent.com/Comfy-Org/workflow_templates/main/input/transparent_rgb_gaming_mouse.png)
>
> Download the default input image, or use your own image.

### Model downloads

> [card] [Diffusion Model: minimax\_h3\_fl2va\_pruned\_int8\_convrot](https://huggingface.co/Comfy-Org/MiniMax-H3/blob/main/diffusion_models/minimax_h3_fl2va_pruned_int8_convrot.safetensors)
>
> Place in `ComfyUI/models/diffusion_models/`

> [card] [Text Encoder: qwen3vl\_32b\_minimax\_h3\_nvfp4\_awq](https://huggingface.co/Comfy-Org/MiniMax-H3/blob/main/text_encoders/qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors)
>
> Place in `ComfyUI/models/text_encoders/`

> [card] [VAE: minimax\_h3\_video\_vae\_int8\_convrot](https://huggingface.co/Comfy-Org/MiniMax-H3/blob/main/vae/minimax_h3_video_vae_int8_convrot.safetensors)
>
> Place in `ComfyUI/models/vae/`

> [card] [VAE: minimax\_h3\_audio\_vae\_fp32](https://huggingface.co/Comfy-Org/MiniMax-H3/blob/main/vae/minimax_h3_audio_vae_fp32.safetensors)
>
> Place in `ComfyUI/models/vae/`

> [card] [Turbo LoRA: minimax\_h3\_fl2v\_turbo\_8step\_v1.0\_comfyui\_bf16](https://huggingface.co/lightx2v/Minimax-h3-Turbo/blob/main/minimax_h3_fl2v_turbo_8step_v1.0_comfyui_bf16.safetensors)
>
> Place in `ComfyUI/models/loras/`. Used when `turbo_mode` is enabled.

### Model storage

```
ComfyUI/
├── 📂 models/
│   ├── 📂 diffusion_models/
│   │   └── minimax_h3_fl2va_pruned_int8_convrot.safetensors
│   ├── 📂 text_encoders/
│   │   └── qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors
│   ├── 📂 loras/
│   │   └── minimax_h3_fl2v_turbo_8step_v1.0_comfyui_bf16.safetensors
│   ├── 📂 vae/
│   │   ├── minimax_h3_video_vae_int8_convrot.safetensors
│   │   └── minimax_h3_audio_vae_fp32.safetensors
│   └── 📂 embeddings/
│       └── minimaxh3_art_is_explosion.safetensors
```

### Prompting tips

1. **Keyframes**: Optional `first_frame` / `last_frame`; motion is generated between them
2. **Prompt**: Shots, motion, and audio in one block
3. **Resolution / duration**: the same rules as T2V (**1344×768** native target, **17k+5** grid)
4. **Turbo**: `turbo_mode` uses `turbo_steps` (default **8**) and `turbo_model_strength` (default **1.0**) instead of 20 steps

`first_frame` is resized to the canvas without keeping its aspect ratio, so an off-aspect image comes out stretched. `last_frame` is center-cropped instead. The I2V template opens at **1:1** and **0.4** MP (**640×640**), so set the aspect ratio to your image and raise Megapixels for a final render.

Official base guide: [prompt guide](https://docs.comfy.org/tutorials/video/minimax/minimax-h3-prompt-guide.md).

## MiniMax H3 Reference to Video (R2V)

Lock character, style, motion, camera, or voice from reference images, videos, and audio.

<video controls src="https://raw.githubusercontent.com/Comfy-Org/workflow_templates/main/output/video_minimax_h3_r2v.mp4" />

> [card] [Run on Comfy Cloud](https://cloud.comfy.org/?template=video_minimax_h3_r2v\&utm_source=docs\&utm_medium=referral\&utm_campaign=minimax-h3)
>
> Open in Comfy Cloud

> [card] [Download Workflow](https://github.com/Comfy-Org/workflow_templates/blob/main/templates/video_minimax_h3_r2v.json)
>
> Download JSON or search "MiniMax H3 R2V" in Template Library

> [card] [Reference Image: red\_superboy\_on\_city\_roof.png](https://raw.githubusercontent.com/Comfy-Org/workflow_templates/main/input/red_superboy_on_city_roof.png)
>
> Character reference for the workflow, or use your own images.

> [card] [Reference Image: mecha\_dragon\_lightning.png](https://raw.githubusercontent.com/Comfy-Org/workflow_templates/main/input/mecha_dragon_lightning.png)
>
> Style and subject reference for the workflow, or use your own images.

### Model downloads

> [card] [Diffusion Model: minimax\_h3\_ref2va\_pruned\_int8\_convrot](https://huggingface.co/Comfy-Org/MiniMax-H3/blob/main/diffusion_models/minimax_h3_ref2va_pruned_int8_convrot.safetensors)
>
> Place in `ComfyUI/models/diffusion_models/`

> [card] [Text Encoder: qwen3vl\_32b\_minimax\_h3\_nvfp4\_awq](https://huggingface.co/Comfy-Org/MiniMax-H3/blob/main/text_encoders/qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors)
>
> Place in `ComfyUI/models/text_encoders/`

> [card] [VAE: minimax\_h3\_video\_vae\_int8\_convrot](https://huggingface.co/Comfy-Org/MiniMax-H3/blob/main/vae/minimax_h3_video_vae_int8_convrot.safetensors)
>
> Place in `ComfyUI/models/vae/`

> [card] [VAE: minimax\_h3\_audio\_vae\_fp32](https://huggingface.co/Comfy-Org/MiniMax-H3/blob/main/vae/minimax_h3_audio_vae_fp32.safetensors)
>
> Place in `ComfyUI/models/vae/`

> [card] [LoRA: minimax\_h3\_ref2v\_turbo\_4step\_v0.1\_comfyui\_bf16](https://huggingface.co/Comfy-Org/MiniMax-H3/blob/main/loras/minimax_h3_ref2v_turbo_4step_v0.1_comfyui_bf16.safetensors)
>
> Required by the workflow's model scan; powers the optional Lightning LoRA turbo mode. Place in `ComfyUI/models/loras/`

### Model storage

```
ComfyUI/
├── 📂 models/
│   ├── 📂 diffusion_models/
│   │   └── minimax_h3_ref2va_pruned_int8_convrot.safetensors
│   ├── 📂 text_encoders/
│   │   └── qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors
│   ├── 📂 vae/
│   │   ├── minimax_h3_video_vae_int8_convrot.safetensors
│   │   └── minimax_h3_audio_vae_fp32.safetensors
│   ├── 📂 loras/
│   │   └── minimax_h3_ref2v_turbo_4step_v0.1_comfyui_bf16.safetensors
│   └── 📂 embeddings/
│       └── minimaxh3_art_is_explosion.safetensors
```

### Prompting tips

1. **Tags**: Reference inputs as `<Picture 1>`, `<Video 1>`, `<Audio 1>` (`ref_image_0` → `<Picture 1>`). Node order: images, then videos, then audio
2. **Assign roles**: Say which reference drives identity, style, motion, camera, or voice
3. **Limits**: Up to **9** images, **3** videos, **3** standalone audio clips
4. **Video FPS**: Reference video is read as **24fps** with no metadata. Convert other frame rates to 24fps before load
5. **Video sound**: Picture and soundtrack are separate references. Wire `ref_video_audio_N` to match `ref_video_N` or the file's audio is not used as `<Audio N>`
6. **Audio numbering**: Video soundtracks get `<Audio N>` first (video order), then standalone clips. Video and picture indices are independent
7. **VAE wiring**: Video soundtrack needs `vae` + `audio_vae`; standalone audio needs `audio_vae` only
8. **Audio roles**: e.g. `use <Audio 1> exactly as it is` vs `follow the timbre of <Audio 2>`
9. **Voice**: Audio carries timbre, not words. Specify delivery; one audio ref per speaker when assignment must be exact
10. **`ref_image_size`**: Default **`match`** (fit to canvas). **`max`** keeps 2048px short edge and is much slower; it does not upscale small files (round sides to 32). Pre-scale to 2048 short edge when needed
11. **Weights**: R2V uses **`ref2va`**, not the T2V/I2V **`fl2va`** model
12. **Turbo**: Optional **4**-step Lightning LoRA (faster, weaker reference/audio). Prefer **20–25** steps when references must hold
13. **Style LoRAs**: Compete with references; use low strength (\~**0.5**) on base schedule when adherence matters

Official ref guide: [prompt guide](https://docs.comfy.org/tutorials/video/minimax/minimax-h3-prompt-guide.md).

## Advanced workflows with the native nodes

Requires a recent ComfyUI build for the features below.

### Anchoring guides at any frame (#15439)

[`MiniMaxH3AddGuide`](https://github.com/Comfy-Org/ComfyUI/pull/15439) anchors a guide at any `frame_idx` (negative = from end), not only first/last frame.

Wire `positive` and latent from an H3 node into `MiniMaxH3AddGuide`. Inputs:

* **image**: Still or clip (multi-frame batches crop to valid **17k+5** lengths; \<5 frames uses first image only)
* **audio**: Soundtrack at the same frame index, cropped to remaining duration
* **frame\_idx**: Anchor position

Connect `vae` for image guides and `audio_vae` for audio. Chain nodes for multiple anchors.

Example: first **22** frames of a clip at frame **0** continues both streams; a still at frame **60** forces the shot through that pose.

This is conditioning, not video-to-video: sampling still starts from the workflow empty latent. `denoise` does not scale guide strength like V2V. To restyle a whole clip, use R2V; to edit regions, use [Fun ControlNet Union](https://docs.comfy.org/tutorials/video/minimax/minimax-h3-fun-controlnet.md) masks.

See the [Multiframe Reference](https://docs.comfy.org/tutorials/video/minimax/minimax-h3-multiframe.md) template.

### Inpainting and extension with latent noise masks (#15375)

[Per-token noise masks](https://github.com/Comfy-Org/ComfyUI/pull/15375) on video and audio latents let you regenerate part of a clip.

Connect a mask to the sampler `denoise_mask`: **0** = keep, **1** = regenerate. Video masks snap to the 2×2 latent patch grid; audio masks to whole latent frames. Use for inpainting, removal, or extension while holding the rest stable.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.