MiniMax H3 Fastest Workflow | 6GB VRAM 16GB RAM — v3.0 — v9.0 (8 Step HD Quality)

This MiniMax H3 workflow turns an image reference into video or transforms an existing video, and includes a video-upscaling option.

It lists image and video loaders, a video saver, and an audio-decoding node.

The listed release is v9.0 (8 Step HD Quality), available from Civitai: https://civitai.com/models/2835250?modelVersionId=3316299.

The workflow lists minimax_h3_ref2va_pruned_int8_convrot.safetensors, minimax_h3_audio_vae_fp32.safetensors, and minimax_h3_video_vae_fp16.safetensors.

It also lists qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors and taeh3.safetensors.

The listed node packs are comfyui_nvidia_rtx_nodes, comfyui-kjnodes, comfyui-spectrum-minimax-h3, comfyui-videohelpersuite, and plaguekind-nodes.

Suggested settings: 4 steps for faster generations, 6 for a balance of quality and consistency, and 8 for the strongest quality and face consistency.

Suggestion · not verified

The update notes say Spectrum is off by default and can be enabled experimentally at 8 steps; they recommend Step 600 Lighting LoRA and describe Tomato Lighting LoRA as an optional stronger-lighting setting for NSFW generations.

For full face consistency, the summary names Eros beta 4; the update also advises keeping the selected models and LoRAs for consistency.

If audio issues arise, try disabling LoRA.

Suggestion · not verified

No download URL is listed for taeh3.safetensors, so check how you’ll obtain that file before using the workflow.

Suggestion · not verified