MiniMax H3 LoRA ComfyUI Workflow (text-to-video + LoraLoaderModelOnly)
======================================================================

Article: https://freakyjolly.com/best-minimax-h3-loras/

What's inside:
  - MiniMax_H3_T2V_with_LoRA.json: ComfyUI workflow (UI format). Drag it onto the canvas.
  - MiniMax_H3_T2V_with_LoRA_api.json: Same graph in API format.

Required models (download each file and put it in the folder shown):
  - minimax_h3_fl2va_pruned-w4a8_convrot_pruned.safetensors
      folder: ComfyUI/models/diffusion_models/
      download: https://huggingface.co/Winnougan/MiniMax-H3-INT4_Convrot_ComfyUI/resolve/main/minimax_h3_fl2va_pruned-w4a8_convrot_pruned.safetensors
  - minimax_h3_turbo_4step_ema_ckpt500.safetensors
      folder: ComfyUI/models/loras/
      download: https://huggingface.co/larryvrh/MiniMax-H3-Turbo-Lora/resolve/main/minimax_h3_turbo_4step_ema_ckpt500.safetensors
  - qwen3vl_4b_int4_convrot.safetensors
      folder: ComfyUI/models/text_encoders/
      download: https://huggingface.co/Merserk/qwen3vl-4b-int4-convrot/resolve/main/qwen3vl_4b_int4_convrot.safetensors
  - mmh3-4b-ClipProj-v3.1-mlp.safetensors
      folder: ComfyUI/models/clip_projections/ (created by ComfyUI-ClipProj on first start)
      download: https://huggingface.co/NicoLab28/ClipProj-MiniMax-H3/resolve/main/mmh3-4b-ClipProj-v3.1-mlp.safetensors
  - minimax_h3_video_vae_int8_convrot.safetensors
      folder: ComfyUI/models/vae/
      download: https://huggingface.co/Comfy-Org/MiniMax-H3/resolve/main/vae/minimax_h3_video_vae_int8_convrot.safetensors
  - minimax_h3_audio_vae_fp32.safetensors
      folder: ComfyUI/models/vae/
      download: https://huggingface.co/Comfy-Org/MiniMax-H3/resolve/main/vae/minimax_h3_audio_vae_fp32.safetensors
  - h3-realism-people-t2v-i2v-r2v.safetensors
      folder: ComfyUI/models/loras/
      download: https://huggingface.co/fal/MiniMax-H3-Realism-People-LoRA/resolve/main/h3-realism-people-t2v-i2v-r2v.safetensors

Custom nodes:
  - ComfyUI-MiniMax-H3-Turbo (MiniMaxH3TurboLoRA, MiniMaxH3TurboSampler): https://github.com/Larryvrh/ComfyUI-MiniMax-H3-Turbo  (ComfyUI Manager > Install Missing Custom Nodes, or git clone into ComfyUI/custom_nodes/)
  - ComfyUI-ClipProj (ClipProjLoader): https://github.com/nicolab28/ComfyUI-ClipProj  (ComfyUI Manager > Install Missing Custom Nodes, or git clone into ComfyUI/custom_nodes/)
  - ComfyUI-KJNodes (DiffusionModelLoaderKJ, PathchSageAttentionKJ, ModelPreviewOverrideKJ): https://github.com/kijai/ComfyUI-KJNodes  (ComfyUI Manager > Install Missing Custom Nodes, or git clone into ComfyUI/custom_nodes/)
  - ComfyUI-Easy-Use (easy cleanGpuUsed, easy clearCacheAll): https://github.com/yolain/ComfyUI-Easy-Use  (ComfyUI Manager > Install Missing Custom Nodes, or git clone into ComfyUI/custom_nodes/)
  - ComfyUI-VideoHelperSuite (VHS_VideoCombine): https://github.com/Kosinkadink/ComfyUI-VideoHelperSuite  (ComfyUI Manager > Install Missing Custom Nodes, or git clone into ComfyUI/custom_nodes/)

How to use:
  1. Install the custom nodes above and restart ComfyUI.
  2. Download the model files into the folders listed (the extra LoRA goes in ComfyUI/models/loras/, a minimax-h3 subfolder is fine).
  3. Drag MiniMax_H3_T2V_with_LoRA.json onto the canvas.
  4. The "Extra H3 LoRA" node sits between the Turbo LoRA and the rest of the model chain. Pick any H3 LoRA from the article there and set its strength from the table; chain more LoraLoaderModelOnly nodes to stack LoRAs.
  5. Put the LoRA trigger word at the start of the prompt (r34l1sm for the fal Realism People LoRA) and queue.

Notes:
  - The example LoRA is fal MiniMax H3 Realism People (single file for T2V/I2V/R2V, strength 1.0, or 0.6 to 0.8 for a lighter touch).
  - The base here is the pruned FL2VA model. A LoRA trained on the full base can misfire on a pruned base, so match pruned LoRAs with pruned bases where the repo ships both.
  - PTD, RAVEN Streaming and the Acc LoRAs need their own loader nodes (see the article), so they are not loaded with LoraLoaderModelOnly.
  - This exact LoRA + Turbo combination has not been benchmarked by me.

Model weights are not included. Each model keeps its own licence; check the model page before commercial use.
Workflow shared by FreakyJolly.com.
