SoL-Refiner
Collection
Speed-of-Light Refinement for High-Resolution Video https://nvlabs.github.io/Sana/Sol-Refiner/ • 5 items • Updated • 9
Open-H3-Regenerate-2K LoRA
The standalone refiner LoRA of Open-H3-Regenerate-2K, trained on the MiniMax-H3 transformer. It regenerates a low-resolution video draft at 2560×1440 in three refinement steps.
Project page · Code · Full release
File
| File | Content |
|---|---|
open_h3_regenerate_2k_refiner_lora.safetensors |
Refiner LoRA, rank 384 on 312 linears of the MiniMax-H3 transformer (Diffusers layout), FP32 |
The inference strength is already folded into the weights: apply the LoRA at scale 1.0.
Keys follow transformer.<module>.lora_A.weight / lora_B.weight.
The full Open-H3-Regenerate-2K recipe also uses the LightX2V Turbo and LMS Sharp LoRAs and a latent upscaler. The full release provides all of them together with the inference code.
License
The refiner LoRA follows the MiniMax H3 Community License, including its territory, use and redistribution terms.
- Downloads last month
- 362
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for Efficient-Large-Model/Open-H3-Regenerate-2K-LoRA
Base model
MiniMaxAI/MiniMax-H3