rzgar/minimax_h3_fl2v_lightx2v_4step_int8-convrot_comfy

🤗 On Hugging Facetext-to-videoapache-2.09.8 GBotherChecksums witnessedupdated today
Magnet

MiniMax H3 FL2V LightX2V 4-Step Int8-ConvRot

_Clip 1: v1 4step | Clip 2: v1 8step_

Your browser does not support the video tag.

_v0.1 4step demo_

Your browser does not support the video tag.

Model description

Requirements

  • ComfyUI (recent - tested on master 0.31.0)
  • ComfyUI-LoraInt8Loader custom node (standard ComfyUI LoRA loaders cannot dequantize it)

Files

| File | Size | NDownload |

|---|---|---|

| minimax_h3_fl2v_turbo_4step_v1.0_int8_convrot | 991 MB | download|

| minimax_h3_fl2v_turbo_8step_v1.0_int8_convrot | 991 MB | Download |

| minimax_h3_fl2v_turbo_4step_v0.1_int8_convrot | 991 MB | Download | 0.99 GB | Full LoRA |

norefiner variant

The full LoRA also patches the model's token_refiner - the text/audio

conditioning refiner. The norefiner variant removes those 8 patches so the

conditioning path runs at base quality (_some video clips had bad audio. This solution fixed it, but keep in mind it doesn’t happen always_).

If audio/text conditioning feels weak with the full LoRA at 4 steps, try the norefiner variant.

_First clip: with refiner | last clip: norefiner_

Your browser does not support the video tag.

Usage (ComfyUI)

1. Unzip the ComfyUI-LoraInt8Loader custom node into custom_nodes, restart

2. In your MiniMax H3 workflow, add LoRA Loader (Int8-ConvRot), select the

file, strength 1.0 (the training scale is already baked in)

3. Sample at 4-6 steps

Download model

Download them in the Files & versions tab.