MiniMax-H3 notebook?

[dkspwndj]

It’s a newly released video model, but it’s a bit much to run locally. Could I ask for a laptop that can run on the Colab A100 model?

[Isi-dev]

It’s a newly released video model, but it’s a bit much to run locally. Could I ask for a laptop that can run on the Colab A100 model?

I don't know if I fully inderstand your question, but google colab runs entirely in the cloud, so you don't need a high end laptop to use an A100 GPU. Any laptop with a web browser and a colab subscription will work fine. I plan to create a comfyUI based colab notebook for MiniMax-H3 soon to make running it easier.

[Isi-dev]

Notebook created: https://colab.research.google.com/github/Isi-dev/Google-Colab_Notebooks/blob/main/Contributions/Minimax_H3.ipynb
I was able to generate a 5 second video with the Q4_K_M GGUF model on the L4 GPU.

[randomhf398pdhg3]

Hey Isi thanks for the great list, can you possibly make a workflow for the pruned int8 version of MiniMax?

There's also a 2nd method explained here (workflow included): https://www.youtube.com/watch?v=VsYm9EYe560

I think they can even be run in T4 as they need 12GB Vram only tho there might be a ram bottleneck. This thing is just 2x better than Wan and LTX in terms of movement and transition.

[Isi-dev]

Hey Isi thanks for the great list, can you possibly make a workflow for the pruned int8 version of MiniMax?

There's also a 2nd method explained here (workflow included): https://www.youtube.com/watch?v=VsYm9EYe560

I think they can even be run in T4 as they need 12GB Vram only tho there might be a ram bottleneck. This thing is just 2x better than Wan and LTX in terms of movement and transition.

I will consider making one for the pruned int8 version. I haven't done any test on the T4 but I think lower GGUF versions of the models should work on the T4.

[randomhf398pdhg3]

That would be very nice, thank you

[Isi-dev]

Hey Isi thanks for the great list, can you possibly make a workflow for the pruned int8 version of MiniMax?

There's also a 2nd method explained here (workflow included): https://www.youtube.com/watch?v=VsYm9EYe560

I think they can even be run in T4 as they need 12GB Vram only tho there might be a ram bottleneck. This thing is just 2x better than Wan and LTX in terms of movement and transition.

The pruned int8 versions of MiniMax-H3 have been included in the UNet / Main Models dropdown in the notebook. You can use the minimax-h3 workflows in the comfyUI templates library or your own workflows and change the models to the ones downloaded by the notebook.