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.
What changed in ComfyUI 0.39?
The official release notes highlight MiniMax-H3 improvements including support for the LynnReal light VAE, an optional VAE path with text-encoder-only references, and LoRA support for MiniMax-H3 and HiDream O1 models trained with DiffSynth-Studio or ModelScope. ComfyUI also changed the default Save Video encoding toward higher quality.
Other useful changes include a fix for int8/int4 cache selection crashes in Qwen Image 2.1, lower peak VRAM and RAM use in the Trellis workflow, more resilient model-asset scanning, and a smaller portable installation.
Why MiniMax H3 users should care
MiniMax H3 workflows have been evolving quickly. The 0.39 changes reduce friction around VAE configuration and broaden LoRA compatibility. 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, Trellis, 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
| Area | What changed | Who benefits |
|---|---|---|
| MiniMax H3 | LynnReal light VAE support, optional VAE/text-encoder-only references, Fun-ControlNet-Union 2.0, VAE fixes | H3 video workflow users |
| LoRA | MiniMax-H3 and HiDream O1 LoRAs trained with DiffSynth-Studio / ModelScope | Fine-tuned workflows |
| Video | Higher-quality Save Video encoding defaults | Video creators |
| Memory | Lower peak VRAM/RAM in TRELLIS workflows | Local GPU users |
| Qwen Image 2.1 | Fix for int8/int4 cache-selection crashes | Quantized 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, TRELLIS, or video export. If your current production graph is stable and does not depend on these areas, there is no SEO-style reason to upgrade just because the version number changed. 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, LoRA compatibility, Fun-ControlNet-Union 2.0 and crash fixes.
Does ComfyUI 0.39 reduce VRAM use?
The release notes specifically report lower peak VRAM and RAM for the TRELLIS workflow. That should not be interpreted as a universal VRAM reduction for every model.
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.
