MiniMax H3 Workflows (T2V, I2V, First-Last-Frame, Ref2V, Lipsync) with SageAttention — v1.1 (faster+upgraded) — v1.5 (turbolora+prompten)

This MiniMax H3 ComfyUI workflow set is for making videos from text prompts or images, with first/last-frame and reference-to-video paths; it takes prompts or source media and outputs video.

The ZIP contains eight workflows: a Base Prompt generator, FL2VA tests with and without Larryvrh Turbo Lora, I2V tests with and without Prompt Enchancer, an R2V Prompt generator, and R2V tests with and without Prompt Enchancer.

The graphs include multiline text prompts and still-image loading; reference-to-video examples also include video and audio uploads, alongside video-combine and audio-decode nodes.

The description points to MiniMax H3 model files and Heretic text encoder options on Hugging Face in INT8 and NVFP4 formats.

Model files named across the workflows include `Heretical-Qwen3.5-9B-fp8.safetensors`, `minimax_h3_audio_vae_fp32.safetensors`, `minimax_h3_fl2va_int8_convrot.safetensors`, and `minimax_h3_turbo_v4_step600_ema.safetensors`.

Other listed model files are `minimax_h3_video_vae_fp16.safetensors`, `minimax_h3_video_vae_int8_convrot.safetensors`, and `qwen3vl_32b_minimax_h3_ultra_uncensored_heretic_int8_convrot.safetensors`.

The graphs name these ComfyUI node packs: `comfy-image-saver`, `comfyui-easy-use`, `comfyui-rmbg`, `comfyui_steudio`, `comfyui-kjnodes`, `comfyui-lora-manager`, `comfyui-minimax-h3-turbo`, `comfyui-pixaroma`, `comfyui-spectrum-minimax-h3`, `comfyui-videohelpersuite`, `rgthree-com

This release is labeled v1.5 (turbolora+prompten); its description says Turbo LoRA workflows use 8 steps and that prompt-enhancer workflows were added.

Hardware notes are not listed for these workflows, so confirm the hardware needs for your chosen path before use.

Vorschlag · nicht geprüft

Some graphs list `comfyui_nvidia_rtx_nodes` at `0.1.3`; check that your ComfyUI setup can provide that exact node pack before relying on it.

Vorschlag · nicht geprüft