Minimax H3 T2V / I2V / REF2V Advanced Filmmaking Workflow | All Speedups + QoL Features — v1.5 — v1.8

This MiniMax H3 ComfyUI workflow produces video clips from text prompts, images, audio, and reference video, and supports T2V, I2V, FFLF, and REF2V modes, video editing, and lip-sync to custom dialogue or music.

Consider FFLF for extending video, or reference video when you want to copy motion into an edit.

Suggestion · non vérifié

Togglable reference fields support up to four pictures, two audio references, and one video reference.

Film VFI interpolation is listed for 24 fps to 48 fps, with an optional temporal upsampler second pass.

The custom image loader's crop and maximum-megapixel controls require `comfyui-obvpm`; the description lists installing it in `custom_nodes` or through ComfyUI Manager's Install via Git option.

The author recommends CUDA 13.0 and says Sage Attention must be reinstalled after a CUDA update.

Model files listed for the workflow are `minimax_h3_audio_vae_fp32.safetensors`, `minimax_h3_fl2va_pruned_int8_convrot.safetensors`, `minimax_h3_ref2va_pruned_int8_convrot.safetensors`, `minimax_h3_video_vae_fp16.safetensors`, and `qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors`.

Node packs listed for the workflow are `comfyui-easy-use`, `comfyui-kjnodes`, `comfyui-videohelpersuite`, and `rgthree-comfy`.

Switch between the T2V/I2V/FFLF model and the REF2VA model in the workflow.

Kijai's Preview Override is the main sampler preview; for non-pixelated previews, the description says to place `taeh3.safetensors` in `/vae_approx/` and set it as the custom VAE.

The author says this workflow has been replaced by a MiniMax Seed Hunter workflow; check that linked workflow before choosing this version.

Suggestion · non vérifié