Video - https://youtu.be/dgts0wG3kh0

Overview

MiniMax H3 is here, and despite the HUGE size, Comfy says it will even run happily on an Nvidia 3060 - https://blog.comfy.org/p/minimax-h3-day-0-support-in-comfyui!

MiniMax H3 is a general-purpose, omni-modal generative system. It supports unified understanding of multimodal contexts composed of text, images, video, and audio, and can generate video with native stereo audio at resolutions up to 2K and durations of up to 15 seconds. Note that the ComfyUI native generation goes up to FULL HD (1920), so alternative upscaling options (e.g. image resize) are needed to go to 2K.

Thanks to its task-generalisation-oriented system design, H3 already possesses broad multimodal context understanding and generation capabilities at the pre-training stage, enabling outstanding performance in following complex multimodal instructions.

H3 supports the following input and output specifications:

Category Specification Output duration 4–15 seconds

Output aspect ratio: Supports a wide range of aspect ratios, including but not limited to 21:9, 16:9, 4:3, 1:1, 3:4, and 9:16

Output resolution: Supports various resolution dimensions. The shorter side is set to 768 pixels by default.

Output frame rate: 24 FPS

Output audio: 32 kHz stereo

Supported dialogue languages: Stable support for 11 languages - Arabic, Chinese, English, French, German, Italian, Japanese, Korean, Portuguese, Russian, and Spanish. Additional languages are also supported to varying degrees

Because I like testing all the things without having to load up a whole bunch of different workflows, I've created a single, combined flow which features:

Text-to-Video

First Frame / Last Frame

Custom Audio Inputs

Custom Video + audio

and more!

MiniMax-H3 features a non-FOSS licence, so be sure to check it out - https://huggingface.co/MiniMaxAI/MiniMax-H3/raw/main/LICENSE

There is also a Q&A about the licence, given that it is so complex: https://huggingface.co/MiniMaxAI/MiniMax-H3/blob/main/docs/QA-about-License.md

For most of the world outside if China, it means filling in a form the get a license from them. I did just that, and the email came through almost instantly.

Workflow Information

To open the workflow, either drag the .json file into Comfyu, or use File -> Open (ctrl-o).

Model Downloads

There are a wide variety of model sizes available, but I've been using the INT8 / NVFP4 versions because they are smaller (and faster)!

Links are provided in-workflow:

Custom Nodes

Only a few custom nodes are required:

Model Loaders

There are 2 MiniMax H3 models - one for References (REF) and one for First/Last frame (FL):

Be sure to select the correct unet_name for each group! The fl2va model should be the one selected in the "Load Model - FL" group:

With the ref2va model being selected in the "Load Model - Ref" group:

Note: I'm using the pruned_int8 versions, but you can use whichever model is best for you!

I found Sage Attention made things a lot faster, but if you don't have that (and Triton) installed, then switch the "Patch Sage Attention KJ" node value of "sage_attention" to disabled and "allow_compile" to false:

You can use either model at the flick of a switch:

First/Last model enabled example:

Ref model enabled example:

Note how the model loader group you aren't using turns purple, which means it is bypassed.

You may wish to start off at 0.2 megapixel resolution to start with to see how things go (this is quite fast and a great way to test prompts):

Video rendering can take a while depending on the length & resolution you use, as well as your own computer hardware.

Prompting

Only a positive prompt is required:

The Qwen3-VL model has decent understanding, so prompts can be quite complex!

Input Options

There are a lot of input options!

Click the switch to toggle between "yes" and "no". If a group is set to "yes" then it will be used. So say, for example, if you only wanted to use a Last Frame (assuming you're using the FL model), then you would only say "yes" to the Last Frame group:

You'll see the nodes in the group(s) you said "yes" to will no longer be coloured purple (aka bypassed). Don't forget to pick the image you want to use as well!

You can change the image size using a number of methods:

Note that the Ref model will scale images anyway, but with First/Last you may wish to change images to match your video resolution.

First and Last Frame notes

If using the FL model then IMG_0 is the First Frame and IMG_1 is the Last Frame. The other images and audio will have no effect. When using the FL model, if the First Frame is enabled, it will always be first, regardless of your prompt - same goes for the Last Frame.

Ref model notes

When using the Ref model then all inputs can be used - the 4 images, video (and audio) & any custom audio too - plus the images can be displayed in any order (you won't always get the First Frame (IMG_0) first)!

Your prompt can make all the difference too. Say for example, like above, you only have IMG_1 enabled.

If you use <Picture 0> in your prompt, you likely won't get the likeness:

However, using <Picture 1> in your prompt instead stands a much better chance (even though the prompt is for a woman!):

You can also increase the number of inputs (by adding more image/video/audio loaders along with set nodes & attaching their get nodes in the engine area) - but there are some limits, as detailed in the note provided:

Inputs

You can use more images if you like, but I've wired up 4 to start with:

There are also 2 x custom video inputs, and 2 x custom audio options. I found on occasion that certain videos without sound could confuse the VHS nodes - but the addition of the Any Switch seemed to prevent such errors. If you do get an error saying that you video doesn't have sound (and so the Video node doesn't want to play ball), simply disconnect the audio output noodle.

The voice cloning works great with 5 second audio clips, but may also work fine with shorter durations. Let me know if you get away with much shorter extracts!

Sampler

You likely won't want to change anything in the sampler other than the random seed:

Despite what Comfy says, I've found the simple scheduler just fine, but experiment away with alternative options should you wish.

Output

The output selection uses the Video Combine node, so you can adjust the various options there, such as for generating PNG or GIF outputs:

Frame Interpolation

Should you wish to go up to 48fps (or more), then enable the Frame Interpolation group and download your choice of RIFE/FILM model:

Change the value in the FPS Multiplier for even more FPS goodness.

The Engine

If you do want to play with the engine, it is up and outside of the usual map:

The MiniMax H3 Reference to Video node can take a few more inputs - so if you want more than 4 images, for example, then connect away!

Enjoy! :)