Added AGSoft Minimax H3 Cache Node to get 20% speed boost without much loss of quality (best cache I've ever used)
Runtime profile
Source description
Added AGSoft Minimax H3 Cache Node to get 20% speed boost without much loss of quality (best cache I've ever used)
Added VRAM + RAM Cleaning nodes
UPD on 08.09.2026
Added more references inputs in Ref2video workflow
Added kitchen attention option
UPD on 13.09.2026
Added optional workflow for Face Refine
Hardware and download-size estimates are not available yet.
AI-generated commentary
AI-generated explanation based on source and configuration details. Suggestions are clearly labeled.
This MiniMax H3 ComfyUI workflow set creates videos from text prompts, images, and reference clips, with workflows for text-to-video, image-to-video, first/last-frame, reference-to-video, lipsync, prompt generation, and face refinement.
The archive contains five workflows: Face Refiner V2V, MiniMax H3 Base Prompt generator, MiniMax H3 I2V, MiniMax H3 Ref2V Prompt generator, and MiniMax H3 Ref2V.
The I2V and Ref2V workflows include image or video loading nodes and video-combining nodes, including LoadImage, PixaromaLoadVideo, and VHS_VideoCombine.
The two prompt-generator workflows include text-entry and prompt-generation nodes; Face Refiner V2V is a separate video-refinement workflow.
The description links to MiniMax H3 model files, INT8 and NVFP4 Heretic text encoders, and a ComfyUI build on Hugging Face.
The workflow files name these node packs: comfyui_fearnworksnodes, comfyui-kjnodes, comfyui-pixaroma, comfyui-videohelpersuite, comfyui_steudio, and comfyui-AGSoft.
They also name comfyui-lora-manager, comfy-image-saver, comfyui-easy-use, comfyui-rmbg, rgthree-comfy, and comfyui_nvidia_rtx_nodes.
Model filenames listed across the workflows include face_yolov8m.pt, gemma4_12b_it_uncensored_heretic_int8_convrot.safetensors, minimax_h3_audio_vae_fp32.safetensors, minimax_h3_fl2v_turbo_4step_v1.2_768p_comfyui_bf16.safetensors, minimax_h3_fl2va_int8_convrot.safetensors, minimx
The title names SageAttention, and version v2.2 adds the AGSoft MiniMax H3 Cache Node.
Update notes mention a kitchen attention option, VRAM and RAM cleaning nodes, more Ref2V reference inputs, and an optional Face Refine workflow.
The listed download is a workflow ZIP, and the description points to model downloads separately; prepare the model files for ComfyUI before running.
Suggestion · not verified
Check the install source for comfyui_nvidia_rtx_nodes and locate model files whose names appear without individual download links.
Suggestion · not verified
Does this need to be edited?
Sign in to send an edit request.
Sources
1 sourceSource excerpts
1 excerptSource context: 9559 downloads · Type ComfyWorkflows · Base model MiniMax H3
Requirements
face_yolov8m.pt
Not resolvedObject Detector · Unknown
minimax_h3_audio_vae_fp32.safetensors
File unverifiedVAE · SAFETENSORS · Unknown
minimax_h3_fl2v_turbo_4step_v1.2_768p_comfyui_bf16.safetensors
Not resolvedLoRA · Unknown
minimax_h3_fl2va_int8_convrot.safetensors
Not resolvedCheckpoint · Unknown
minimax_h3_video_vae_int8_convrot.safetensors
Not resolvedVAE · Unknown
person_yolov8m-seg.pt
Not resolvedObject Detector · Unknown
qwen3vl_32b_minimax_h3_ultra_uncensored_heretic_int8_convrot.safetensors
Not resolvedText encoder · Unknown
Node pack · Registry
Node pack · Registry
comfyui-pixaroma
Node pack · Registry
comfyui-videohelpersuite
FoundNode pack · Registry