Introduction

ComfyUI v0.41.0 is tagged and live. The GitHub Release landed at 2026-10-15T22:57Z (about 06:57 MYT Oct 16), just two days after v0.40.0, and it is short but very on-trend: native FLUX.3 Image Klein support, a brand-new Convert SDR to HDR node, a friendlier multi-reference latent node for edit models, and MiniMax H3 LightVAE loading. It is open-source, runs locally on your own GPU, and ships with fresh Windows portables today.

Release: ComfyUI v0.41.0 · Compare: v0.40.0…v0.41.0

Grab the build

Windows portables are attached to the tag:

  • ComfyUI_windows_portable_nvidia.7z
  • ComfyUI_windows_portable_nvidia_cu126.7z
  • ComfyUI_windows_portable_nvidia_arm64.7z
  • ComfyUI_windows_portable_amd.7z
  • ComfyUI_windows_portable_intel.7z

Git installs can pin the tag directly:

cd ComfyUI
git fetch --tags
git checkout v0.41.0
pip install -r requirements.txt

FLUX.3 Image Klein goes native

This is the headline. Black Forest Labs dropped FLUX.3 Image Klein as 5.6B open weights yesterday, and v0.41.0 is the first tagged ComfyUI release that carries the core model code (#17064, by kijai).

  • Two checkpoint flavours: the base model and the step-distilled "flow map" checkpoint.
  • Recommended settings from the PR: flow map uses euler, 16 steps, CFG 1; base uses 50 steps, CFG 3.5.
  • Text encoder: stock Qwen3-VL-4B, loaded through CLIPLoader with type flux3.
  • VAE: the existing Flux2 VAE, so you may already have it on disk.
  • New FLUX.3 Layout Prompt node: takes boxes from Create Bounding Boxes and appends them to your prompt as FLUX.3's layout table (obj_N ids on a 0–1000 grid). That is Klein's bounding-box control, wired straight into the graph.

The bundled workflow templates jump to v0.11.84, which adds four ready-made Klein graphs: image_flux3_klein_t2i, image_flux3_klein_image_edit, and distilled versions of both. The same template drop also adds api_krea3_t2i and api_krea3_image_edit for the Krea 3 partner nodes (more on that below).

Pro Tip

Start with the distilled template. Sixteen steps at CFG 1 is the quick-iteration path; switch to the base checkpoint at 50 steps when you are ready for the final render.

New native nodes

  • Convert SDR to HDR (ImageSDRToHDR, #16990): expands ordinary sRGB images into Rec.2020 HLG HDR. It ships with one algorithm, lottes, and a target_luminance control that defaults to 650 nits (range 0–10,000). Alpha is preserved. Save the result with the HDR color space selected, and pair it with the EXR save improvements from v0.40.0.
  • Set Reference Multi Latent (ReferenceMultiLatent, #17009 and #17053): encodes reference images through your VAE and appends their latents to positive (and optional negative) conditioning for edit models. Inputs auto-grow up to 100 image slots, and you can keep each reference at its native size or resize it to a megapixel target (1.0 is about 1024×1024). It also outputs an empty latent sized from your first reference, which saves a wiring step.
  • Latent geometry nodes (#16093): read latent sizes for overlap maths and noise-mask setups, with separate video and audio latent nodes so nested audio-video latents stay clearly labelled.
  • Default workflow route (#16956): a core route for the default workflow, which matters if you build tools on top of ComfyUI.

MiniMax H3 LightVAE support

#17049 adds loading for lightx2v/LightVAE on MiniMax H3, plus the option to load the VAE encoder and decoder separately. We covered why LightVAE matters for faster H3 and Wan decoding in our LightVAE Dispatch; now it is in a pinned release instead of master.

Same-day companions

Krea 3 Partner Nodes. Hours before the tag, ComfyUI announced Krea 3 in Partner Nodes: reference-driven generation and editing with multiple references addressed by upload order. These are paid API nodes, and Krea 3 itself is still in beta, so read our Krea 3 early-access breakdown for specs and pricing.

A $25,000 open-source build challenge. Comfy also launched a two-week competition for the best open-source apps built on the Comfy Developer Platform, pitched as open rebuilds of Higgsfield's Genjutsu, Cinema Studio, or anything new. Per the Comfy blog post:

  • Grand Prize: $10,000 for the best overall build
  • Most Innovative, Most Viral, and Open Weights Only: a GeForce RTX 5090 each
  • 10 runners-up: $250 in Comfy Credits each
  • Partner bonuses: $2,000 cash for an LTX-core VFX workflow, $1,000 / $500 / $250 in FLUX API credits for FLUX3-core builds, and $1,000 in MiniMax API credits for H3-core builds

Backers listed are NVIDIA AI, Runpod, LTX, Black Forest Labs, and MiniMax, with in-person build events in SF (Oct 28), LA (Oct 29), and NYC. Fresh native Klein and H3 LightVAE support make the "Open Weights Only" track a lot more tempting.

Original Source

Conclusion

ComfyUI v0.41.0 is a small tag with big timing: FLUX.3 Klein lands natively with layout-box prompting and four starter templates, SDR images can now become HLG HDR inside the graph, multi-reference editing gets a proper node, and H3 LightVAE is finally pinned. Grab a portable or check out the tag, load the distilled Klein template, and maybe start sketching your challenge entry while you are in there.

Prefer to skip the install? Try FLUX on Gen → https://artrealmai.com/gen?utm_source=magazine&utm_campaign=gen&utm_content=comfyui-v0-41-0-release

—Aurelia ♡