ComfyUI v0.45.0 came out on 10 November 2026, six days after v0.44.0. It is a big one for local users: native support for MiniMax-H3-Image and the RelayVSR video upscaler, plus WeVisOne-VL, UltraTex and a pile of new partner (API) nodes.
Here is everything that changed, what it means for you, and what to check after updating.
- Released: 10 November 2026 (v0.44.0 was 4 November).
- New local models: MiniMax-H3-Image (16 MP images and editing), RelayVSR (fast 2x/4x video upscaling), WeVisOne-VL and UltraTex.
- New partner nodes: Pika H3 Flex and audio, Kling 4.0 Flash, GPT Image 2.6 and V3, Gemini Omni Flash 1.2, Seedream 5.5, Seedance 2.5 at 4K, Tencent HY LLM, Breeze TTS 2.
- Memory wins: adaptive SeedVR2 VAE slicing and Get Video Components now keeps decoded frames once instead of twice.
- Check after updating: the run_cpu and portable fast fp16 scripts moved to
./advanced, and Qwen Image nodes moved to Comfy Router.
New Local Models
These run on your own GPU with downloadable weights.
| Model | What it does | PR | Our guide |
|---|---|---|---|
| MiniMax-H3-Image | 33.1B image model, about 16 MP output, text to image and editing with up to 16 references in one checkpoint | #17655 | H3-Image setup guide |
| RelayVSR | Fast 2x or 4x video upscaler that runs the heavy model on every 15th frame only | #17511 | RelayVSR guide |
| WeVisOne-VL | Vision-language model support | #17632 | WeVisOne-VL guide |
| UltraTex | 3D texturing, with texture rebaking | #17636 | Not covered yet |
MiniMax-H3-Image is the headline. It renders 4096 x 4096 in six steps at CFG 1 and edits images by naming references as <Picture 1>, <Picture 2> and so on. The files are big (about 30 GB at minimum), and it needs template pack 0.11.104 or newer.
RelayVSR was already on the master branch since 5 November. This is the first stable release with its RelayVSR Conditioning and VSR Decode nodes. LightX2V’s H100 test put X2 at 45 FPS, against 0.69 FPS for SeedVR2-3B.
New Partner (API) Nodes
Partner nodes run on the provider’s servers and use Comfy credits. No model downloads are needed.
- Pika: H3 Flex nodes (#17509), plus Soundtrack, SFX and Music nodes (#17328).
- Kling: the Kling 4.0 Flash model, and an original-audio option for Kling 4.0 nodes (#17536).
- OpenAI: GPT Image 2.6 models (#17558), and GPT Image V3, GIF and SVG nodes (#17591).
- Google: Omni Flash 1.2 in the Gemini Omni video node (#17560).
- ByteDance: Seedream 5.5 (#17631) and 4K output for Seedance 2.5 (#17579).
- Tencent: an HY LLM node (#17534).
- Breeze: TTS 2 voice and text-to-speech nodes (#17611).
Grok prices and tooltips were also updated for retired image models (#17532). As always, check the price badge on each partner node before you run it.
If you missed last week’s MiniMax H3 Max editing nodes from v0.44.0, our H3 Max editing nodes guide covers all six.
Speed and Memory Improvements
- SeedVR2 VAE: the slice length now adapts to free memory (#17508), which should mean fewer out-of-memory errors when upscaling video on smaller cards.
- Get Video Components: decoded frames are held once instead of twice (#17562). Long clips use noticeably less RAM.
- Qwen 3.x text models: video input support (#17582), combined parallel linear weights (#17361) and better prefetch handling (#17625).
- Text generation: repetition and presence penalties now apply in greedy decoding (#17635).
- Mesh: a Solid mode for Extract Mesh from Splat (#17497).
Bug Fixes
- A potential crash (#17543) and a crash in UNetSelfAttentionMultiply (#17603).
- A Qwen fp8 crash during text generation (#17620).
- A potential database issue (#17602).
- Remesh on imported meshes (#17531).
- Megapixel precision in the resolution selector (#17555).
- A clearer error when a pixel-space VAE is picked by mistake (#17518). If you have ever had a cryptic shape error from the wrong VAE, this one helps.
Changes That Might Affect You
No breaking changes are listed, but a few things moved:
| Change | What to do |
|---|---|
run_cpu and portable fast fp16 scripts moved to ./advanced (#17571, #17570) |
If you start ComfyUI Portable with one of these .bat files, or have a shortcut to one, run it from the new advanced folder |
| Qwen Image nodes moved to Comfy Router (#17527) | If a saved Qwen Image workflow shows missing nodes, search for them again by name or reload the latest template |
local_web_sso and partner_node_token on by default (#17540) |
These appear to relate to signing in for partner nodes. Most local-only users should not need to change anything |
| Frontend 1.58.5, comfy-kitchen 0.2.41, templates up to 0.11.105 | Update the frontend and templates packages along with the core |
How to Update
- Back up your
custom_nodeslist and any workflows you care about. - ComfyUI Desktop: it offers the update itself. Accept it and restart.
- Portable: run
update_comfyui.batin theupdatefolder. - Git install: in the ComfyUI folder, run
git pull, thenpip install -r requirements.txtin the same Python environment. - Restart and hard-refresh the browser (Ctrl + F5).
Our full ComfyUI update guide covers each install type and how to roll back. New to ComfyUI? Start with the ComfyUI beginner guide.
If something breaks, the usual cause is an outdated custom node. Update all nodes in ComfyUI Manager, and see our Manager guide if it is not installed.
FAQ
Is ComfyUI 0.45.0 a stable release?
Yes. It was tagged on 10 November 2026. Comfy Cloud and the Desktop app pick it up after the stable tag.
Do I need 0.45.0 for MiniMax-H3-Image?
Yes. The MiniMax H3 Image Conditioning node is new in this release, along with template pack 0.11.104.
Will my old workflows still work?
Almost all will. Watch for Qwen Image workflows, because those nodes moved to Comfy Router.
Are the partner nodes free?
No. They use Comfy credits and need a signed-in account. Local models stay free.
Where is the official changelog?
On the Comfy docs changelog and the GitHub release page. At the time of writing the docs page still showed v0.44.0 as the newest entry.
Sources
- Freedom.Tech: ComfyUI 0.45.0 release summary (full PR list)
- Comfy blog: MiniMax H3 Image is now available (10 November 2026)
- ComfyUI docs: MiniMax-H3-Image tutorial (version requirements)
- ComfyUI Wiki: RelayVSR X2 (merge date, benchmarks)
- ComfyUI changelog (v0.44.0 date)
