LMS / Sharpness LoRA
minimax_h3_lms_v1.0_r64.safetensors is a
rank-64 LoRA for sharpening a source video while keeping a natural, photorealistic appearance.
It is intended mainly as a second pass over an existing clip.
Workflow
workflows/minimax_h3_lms_workflow.json
Put the source video into MiniMaxH3AddGuide, connect the video VAE, use frame_idx = 0, and
match the output resolution and frame count to the source. Start with LoRA strength 1.0.
Use this caption:
Enhance this video with sharp, crisp details while preserving a natural photorealistic appearance.
The aligned guide is not an ordinary reference video. Its latent frames occupy the same temporal and spatial positions as the output, helping preserve motion, timing, framing and sync.
The model was trained primarily for ref2va with guide latents alone. Combining Add Guide with
additional native references is experimental and may not improve the result.