Skip to content

run-comfyui-minimax

Release Notes Policy

  • A new dated section is added whenever a new stable ComfyUI release is included.
  • The date is updated whenever custom nodes, provisioning, or runtime behavior change.
  • This pod is stable, but many of its custom nodes are changing with every version.
  • Some custom nodes have been adapted specifically for this pod or pinned to increase the stability.
  • Tested on L40S, RTX 6000 Ada, RTX PRO 6000, PRO 6000 MiG 48Gb, RTX 5090, (RTX 4090).

14102026

11102026

  • Updated ComfyUI to version 0.39.0.
  • Updated omnichar previewing image when a char is uploaded.
  • Updated workflows added latent upscaler.
  • ComfyUI-MiniMaxH3Mod.
  • Updated Fantastic-prompt-builder to upload/download refmods in zip format.
  • Lowered container volume.

05102026

  • Updated ComfyUI to version 0.38.0.
  • Updated Fantastic Prompt Builder to automatically download SAM3 when needed for video masking.
  • Updated the H3 Latent Upscaler to automatically download its model.
  • Fixed a crash in Director when using "segment continuity" by switching to external Motion Context.
  • Removed the PDD-Acc workflows for Ref2VA, as Hyperflow delivers better results.
  • ComfyUI-Omnichar.
  • Updated QwenImageRefPack with an extra omnichar load and edit node.
  • Updated Omnichar adding uploading and downloading char files, extending slots to 9 images.

29092026

Container CUDA upgrade notice

  • run-comfyui-minimax requires CUDA 13.x (due to breaking changes in "comfy-kitchen" and no support for CUDA 12.x on modern GPU's by ComfyUI).
  • If you need due GPU shortage a instance change runpod container template tag to run-comfyui-minimax:21092026.

21092026

  • ComfyUI 0.37.0
  • Last tested version running on pytorch 2.10 CUDA 12.8

20092026

  • ComfyUI 0.36.0
  • Removed unused custom nodes.
  • Lowered pod size by excluding unused media in custom_nodes
  • Added GPU-accelerated video encoding with FFmpeg and NVENC to speed up video creation on slower vCPUs.
  • Auto detect NVENC availability on host.
  • Updated worklows to use nvenc video encoding.
  • Forked and customized model_linker.
  • Added another video VAE to reduce memory usage (does not work with pruned models).
  • Updated Lora-Manager template.
  • Forked and fixed VHS : video upload without audio regression.
  • ComfyUI-Hyperflow
  • Added workflows Hyperflow 8 step.
  • comfyui-obvpm

15092026

  • ComfyUI 0.35.0
  • Updated workflows.
  • Tail is an optional download.
  • Switched Blackwell HVRAM diffusion models to standard INT8 ConvRot for better compatibility.

08092026

  • ComfyUI 0.34.0
  • ComfyUI-MiniMax-H3-PDD-Acc
  • Fixed Multi-shot audio crash.
  • Removed custom_nodes using irriversible patching
  • Security updates

25082026

  • ComfyUI 0.33.0
  • Fixed THRESHOLD typo.
  • Updated configuration for RTX PRO 6000 for higher quality.
  • Changed configuration diffusion model LVRAM BACKWELL for lora compatibility.
  • Added local llama-cpp (cli,server) support
  • Added separate RunPod templates for local Qwen-VL prompt enhancement with llama.cpp; see the deployment guide for more information.
  • Added uncensored profile to_QWEN_H3_prompt modified workflows.
  • Fixed crash mult-shot context pin after first shot in workflows (sol attention)
  • ComfyUI-H3-Qwen3VL-TextGen
  • MiniMax H3 Audio T8
  • ComfyUI_Qwen_H3_Prompt

13082026

11082026

07082026