A ComfyUI workflow built around Capricorncd Timeline and MiniMax H3 Video Generator, with an editable timeline and examples of five video generation modes.

Features

• Text to Video (T2V)

• Image / Multi-Image Reference to Video (I2V)

• Video Reference to Video (V2V)

• First & Last Frame to Video (FL2V)

• Audio-driven lip-sync avatars

• Optional two-pass latent upscaling and refinement

• Sampling previews, subtitles, audio tracks and video export

• Optional face refinement and motion deblur — experimental and not extensively tested in this workflow

Required Nodes

• Updated ComfyUI with native MiniMax H3 support

• ComfyUI-Capricorncd-Timeline

• ComfyUI-Easy-Use

• ComfyUI-KJNodes

• Comfyui_Minimax_h3_latent_Upscaler for two-pass upscaling

• facok/comfyui-SelfLift

• FFmpeg for media processing and export

Optional features require ComfyUI-H3-FaceRefine, ComfyUI-MAINodes (motion deblur), ComfyUI-H3-AudioRefine, ComfyUI-WanVideoWrapper (loudness normalization), or ComfyUI-H3-Motion-Context.

Models

Place models in the corresponding ComfyUI folders:

FolderModelmodels/diffusion_modelsMiniMax_H3-Pruned-Ref-Delta-Fused-r1024-comfy-int8-convrot.safetensorsmodels/diffusion_modelsMiniMax_H3_FL2VA_int8_convrot.safetensors for FL2Vmodels/text_encodersqwen3vl_32b_heretic_minimax_h3_nvfp4.safetensors, or a compatible H3 text encodermodels/vaeminimax_h3_video_vae_fp16.safetensors and minimax_h3_audio_vae_fp32.safetensorsmodels/lorasMatching ref2v or fl2v Turbo LoRAs for the selected 4/8-step sampling modemodels/latent_upscale_modelsminimax_h3_latent_upscaler_3d_fp16.safetensors for two-pass upscalingmodels/vae_approxtaeh3.safetensors for RGB sampling previewsmodels/ultralytics/bboxface_yolov8m.pt for face refinement

Select the appropriate model and matching LoRA before generating each mode. Motion deblur also requires a separate base model without an acceleration LoRA.

How to Use

Load the workflow, open the Timeline Editor, click Import ZIP in the upper-right corner, and select the corresponding project and assets ZIP package from workflows/resoure.

Select your installed models, choose the Clips to generate, and run.

Installation and project repository · MiniMax H3 Video Generator documentation