Downloads · 30 days
0
WarmBloodAban/Minimax_h3_ref2va_XUELUO_int8_convrot
Minimax_h3_ref2va_XUELUO_int8_convrot is a machine learning model from WarmBloodAban. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
minimaxh3ref2vaXUELUOint8convrot is a video generation model deeply optimized with Layer Merging, built upon the INT8 version of the MiniMax-H3 model library.
Downloads · 30 days
0
Access
Public
Updated Aug 16, 2026
Repo size
34 GB
Likes
33
Public
Click a slice to open those files.
.safetensors34 GB · 100%
From the Hugging Face model README
minimax_h3_ref2va_XUELUO_int8_convrot is a video generation model deeply optimized with Layer Merging, built upon the INT8 version of the MiniMax-H3 model library.
Credits: Special thanks to @雪落眉间路已远 for providing the core technical approach. He is currently using this H3 model to create AI short dramas on Bilibili. Check out his work here: Bilibili Space.
This model addresses the key pain points of original video generation models regarding image quality and stability. By combining the visual detail advantages of fl2va with the structural stability of ref2va, it delivers an enhanced Reference-to-Video experience.
fl2va significantly outperforms ref2va in visual detail. This model precisely extracts the core feature layers (Layers 25–49) from fl2va and seamlessly merges them into the ref2va architecture.fl2va with the temporal consistency and subject stability of ref2va, drastically reducing artifacting and distortion.Model-Configured Dual-Sampling Workflow: 👉 https://www.runninghub.ai/post/2088996582058835969/?inviteCode=rh-v1559
| Parameter | Setting |
|---|---|
| Base Model | MiniMax-H3 INT8 |
| Generation Mode | Ref2VA / Reference-to-Video |
| Sampling Steps | 4 – 20 steps (Supports 4-step LoRA ultra-fast rendering) |
| Best Used For | Pairing with high-detail reference images/videos for high-quality, dynamically stable video generation |
Special thanks to the original model creators and the ComfyUI community for their foundational work: