Home/ COMFYUI/ ComfyUI 0.39: What Changed for MiniMax H3 Workflows?
COMFYUI

ComfyUI 0.39: What Changed for MiniMax H3 Workflows?

ComfyUI 0.39 adds important MiniMax H3 improvements, higher-quality video encoding defaults, Qwen Image 2.1 fixes, and workflow memory optimizations. Here is what matters in practice.

Published Oct 7, 2026 · 3 min read
ComfyUI 0.39 update featuring MiniMax H3 workflow improvements

ComfyUI 0.39.0 landed on October 5, 2026, and this release is especially relevant if you are building local AI video workflows around MiniMax H3. Rather than being a cosmetic update, v0.39.0 changes several pieces that affect video generation, memory use, and model compatibility.

Looking for a MiniMax H3 workflow? This article covers verified v0.39.0 release changes; it is not a tested H3 setup tutorial. Confirm model availability, license, required files and hardware in the upstream workflow documentation before choosing a local or API-based setup.

What changed in ComfyUI 0.39?

The official v0.39.0 release notes list support for the LynnReal light MiniMax-H3 VAE, a MiniMax VAE offload fix, and earlier release of packed H3 embedding temporaries. ComfyUI also changed the default Save Video encoding toward higher quality.

Other listed changes include a fix for int8/int4 cache selection crashes in Qwen Image 2.1 and more resilient model-asset scanning.

Why MiniMax H3 users should care

MiniMax H3 workflows have been evolving quickly. The 0.39 changes reduce friction around VAE configuration and adjust embedding memory handling. That matters more than the version number itself: fewer model-specific workarounds make H3 easier to integrate into repeatable ComfyUI pipelines.

Should you update?

If you actively use MiniMax H3, Qwen Image 2.1 or ComfyUI video workflows, v0.39 is worth evaluating. For a production workflow, keep a copy of your working environment and custom-node versions before upgrading, then validate your existing workflows after the update.

How to update or start clean

If you need a clean installation first, follow our ComfyUI installation guide for Windows, macOS and Linux. For production pipelines, continue with our ComfyUI video workflow guide.

What The Signal will test next

We are tracking MiniMax H3 as part of our practical local-video coverage. We will separate first-party tests from upstream project claims and update this article when we have reproducible workflow results.

ComfyUI 0.39 at a glance

AreaWhat changedWho benefits
MiniMax H3LynnReal light VAE support, VAE offload fix, earlier release of embedding temporariesH3 video workflow users
VideoHigher-quality Save Video encoding defaultsVideo creators
Qwen Image 2.1Fix for int8/int4 cache-selection crashesQuantized image workflows

These are upstream release-note changes, not independent The Signal benchmarks.

ComfyUI 0.39 vs 0.38: should you upgrade?

Upgrade pressure is highest if your workflow uses MiniMax H3, Qwen Image 2.1 or video export. If your current production graph is stable and does not depend on these areas, evaluate whether the documented fixes address a problem you actually have. Back up the working environment and custom-node versions, update, then validate the same workflow and seed before adopting the release in production.

ComfyUI 0.39 FAQ

Does ComfyUI 0.39 improve MiniMax H3?

Yes. The official release includes multiple H3-specific changes covering VAE handling and embedding memory management.

Does ComfyUI 0.39 reduce VRAM use?

The release notes include MiniMax VAE offload and embedding-temporary changes. They do not establish a universal VRAM reduction or a measured saving for your workflow.

Should I update ComfyUI immediately?

For experimentation, the release is attractive. For production, preserve your known-good environment and validate existing workflows after updating, especially custom nodes.

Is MiniMax H3 built into ComfyUI?

ComfyUI has native H3-related support and v0.39 expands it. Exact workflow requirements still depend on the models, VAE, LoRAs and nodes used.

Sources

The Signal newsletter

Keep getting this

One edition a week on open models, local setups and the tools around them.

Read the latest issue

Email delivery opens once the newsletter platform is connected.