# HyperFlow H3 Flash Faster Generation with 6GB of Vram — retained record

> Retained reddit record from reddit:comfyui.

Canonical: https://highsignal.app/data/records/4f0c160a14f3a21d

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


source: [reddit](https://highsignal.app/data/reddit) (reddit:comfyui)  published 2026-11-19 14:26 ingested 2026-11-20 02:39

[original source ↗](https://www.reddit.com/r/comfyui/comments/1y0antl/hyperflow_h3_flash_faster_generation_with_6gb_of/)

## retained text



```
Hello everyone 

I’ve been testing **HyperFlow**, a compressed 14B-parameter open-source model created through self-distillation acceleration of the 20B MiniMax H3 model, and I built a complete ComfyUI workflow around it. The main reason I wanted to test it was speed: it aims to reduce generation time without relying on external training data while still maintaining good quality. It also supports **NVIDIA Sol-Attn sparse attention**, which can provide another speed improvement, especially when combined with the **SelfLift sampler**. 

I also added **RelayVSR** for upscaling the final result. One thing that surprised me is how well it handles **characters and movement** compared with some of the other MiniMax H3 workflows I’ve tested. It doesn’t require a LoRA and works well at only **6 steps**. On my **6GB VRAM** setup, I managed to generate around **0.7 megapixels in about 20 minutes**, which is roughly twice as fast as my previous setup. 

The main drawback I’ve noticed so far is **distorted faces**, especially in long distance prompt some generations. I’m currently testing different ways to fix that. Overall, I think HyperFlow is a very interesting option for anyone trying to run high-quality video generation on limited VRAM. I’ve included the complete workflow on Patreon for those who want to test it themselves.

***Models & Links***

[https://huggingface.co/BennyDaBall/HyperFlow-H3-Flash-INT8-ComfyUI/tree/main](https://huggingface.co/BennyDaBall/HyperFlow-H3-Flash-INT8-ComfyUI/tree/main)

[https://github.com/Video-Rebirth/ComfyUI-HyperFlow-Flash](https://github.com/Video-Rebirth/ComfyUI-HyperFlow-Flash) [https://github.com/BennyDaBall930/comfyui\_hyperflow\_int8\_nodes](https://github.com/BennyDaBall930/comfyui_hyperflow_int8_nodes)

***Workflow Links***

[https://drive.google.com/file/d/1Ov6VGFgWv5EHQ9Dzof0Z9d8irIwG49sX/view?usp=sharing](https://drive.google.com/file/d/1Ov6VGFgWv5EHQ9Dzof0Z9d8irIwG49sX/view?usp=sharing)

[https://civitai.com/articles/39206/hyperflow-h3-flash-faster-generation-with-6gb-of-vram](https://civitai.com/articles/39206/hyperflow-h3-flash-faster-generation-with-6gb-of-vram)

***Video Tutorial Link***

[https://youtu.be/EiIhsiWnKdQ](https://youtu.be/EiIhsiWnKdQ)


```



document fetched 2026-11-20 00:42 · canonical https://reddit.com/r/comfyui/comments/1y0antl/hyperflow_h3_flash_faster_generation_with_6gb_of
