# MiniMax H3 - Really is the best open-source video model. — retained record

> Retained reddit record from reddit:StableDiffusion.

Canonical: https://highsignal.app/data/records/3efdd5d51d0a0ce9

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:StableDiffusion)  published 2026-10-20 03:36 ingested 2026-10-21 02:38

[original source ↗](https://www.reddit.com/r/StableDiffusion/comments/1xahodr/minimax_h3_really_is_the_best_opensource_video/)

## retained text



```
View on YouTube while video is pending here if needed. [https://youtu.be/87w4GFO6fd4](https://youtu.be/87w4GFO6fd4)

I did only a small amount of manual work on this. Barley any post edits besides one filter and one tiny scene edit.

However,  this post [Here](https://www.reddit.com/r/comfyui/comments/1vfg9sf/minimax_h3_first_test_using_it_inside_my_video/) was using only the video builder manually with open source only and no mcp. and it did a one shot music video. just so u can see that can use it with local llms

Keep your rude and disrespectful comments to yourself. They help no one. Polite and respectful feedback is fine but this video turned out exactly as I planned. The expressive facial expressions on the singer were intended since it's supposed to theatrical.

• The song, made with Suno, lyrics created by Chat GPT.

• I Created a GPT plug-in that I was testing. It creates the full band, the song, the song style and the locations. I just gave the song to Suno.

Model settings: MiniMax H3 Singularity, 25 steps on pass 1, then the 8-step 768p turbo LoRa at 4 steps and 0.35 denoise on pass 2 with the latent upscale model.

This used my ComfyUI video builder which has MCP/API support so you can have an agent run it for you and then open up the created project later to see it and play around with it.

The video builder can create all the prompts for you but for this Video I did have Claude use the API / MCP stuff to help create the prompts. You can use any LLM agent to do this.

What Claude did on its own:

• Planned all the scenes and wrote every H3 prompt shot by shot, with per-shot lyric cues and emotion tags

• Rendered overnight through my VRGDG Video Builder's MCP/API.

• Checked every take frame by frame and re-shot the problems.

• Checked every cut, and the audio sync of all the scenes came out at 0 ms

You can find my Video builder on GitHub [https://github.com/vrgamegirl19/comfyui-vrgamedevgirl](https://github.com/vrgamegirl19/comfyui-vrgamedevgirl)  
We are working on a new story mode which will allow you to create video's just like this one when it comes to the story flow - But no need for Claude!

You can find a 2-pass workflow in my workflows folder.

Workflows\\MiniMaxH3\_2Pass\_LatentUpscale\_Shareable.json

I have a discord community and post updates and walkthroughs. If you want to try the builder for yourself come join the fun!

 [https://discord.gg/gmdm9EeD39](https://discord.gg/gmdm9EeD39)
```



document fetched 2026-10-21 00:31 · canonical https://reddit.com/r/StableDiffusion/comments/1xahodr/minimax_h3_really_is_the_best_opensource_video
