LiconStudio/Minimax-H3-VBVR · Hugging Face

Source: https://huggingface.co/LiconStudio/Minimax-H3-VBVR

https://huggingface.co/LiconStudio/Minimax-H3-VBVR#overviewOverview

Licon VBVRis an image-to-video LoRA forMiniMax-H3, trained on VBVR data rebuilt and reprocessed specifically for H3.

It improves object consistency and character action logic, reducing unintended object duplication, clipping, and interpenetration during motion and interactions.

https://huggingface.co/LiconStudio/Minimax-H3-VBVR#key-improvementsKey Improvements

• **Reduced object duplication:**fewer unintended copies of objects during generation.
• **Reduced clipping and interpenetration:**improved spatial interactions between characters, objects, and surfaces.
• **Improved character action logic:**more coherent movement sequences and interactions.

https://huggingface.co/LiconStudio/Minimax-H3-VBVR#training-dataTraining Data

The training data is derived from**VBVR-Pro-SFT-Video**.

I rebuilt the VBVR training data around H3’s characteristics, adapting data preparation and preprocessing to its training pipeline. The goal is to transfer structured motion and interaction patterns into practical video generation, with a focus on object consistency, spatial relationships, and action logic.

https://huggingface.co/LiconStudio/Minimax-H3-VBVR#video-comparisonsVideo Comparisons

Example 02Original MiniMax-H3-FL2VAMiniMax-H3 + VBVRExample 03Original MiniMax-H3-FL2VAMiniMax-H3 + VBVR## https://huggingface.co/LiconStudio/Minimax-H3-VBVR#usageUsage

Download**H3-VBVR-I2V.safetensorsand load it with a compatibleMiniMax-H3 image-to-video workflow**.

For ComfyUI, place the file inComfyUI/models/loras/and select it in the LoRA loader used by your H3 workflow.

https://huggingface.co/LiconStudio/Minimax-H3-VBVR#acknowledgmentsAcknowledgments

Thanks to**MiniMaxfor the base model and theVBVR team**for the source dataset.