ComfyUI News & Open-Source AI Releases

OpenMOSS released MOSS-Image, a 5B text-to-image DiT with layout JSON control and strong text rendering, plus Base and SFT checkpoints and same-day ComfyUI support.

Plenio 1.0 turns ComfyUI into a free local song studio: a language model writes the lyrics, YuE2 or MiniMax Music 3 performs them, and a new browser player plays and shares them.

Comfy Org's Open Source Higgsfield challenge puts $25,000 in prizes behind open-source rebuilds of apps like Genjutsu on the Comfy Developer Platform, open to October 30.

Black Forest Labs released FLUX 3 Image Klein: a 5.6B open weights image model for text-to-image, editing and multi-reference generation, with day-one ComfyUI support.

Alibaba's TaoLive AIGC team released TaoMate-H3-FL2VA, a 3-step LoRA that adds first/last-frame control to streaming MiniMax H3 generation, with day-one ComfyUI conversions.

minimax-h3-community-license-agreement

ComfyUI-Drama3 is a node pack that runs Fish Audio Drama 3 inside ComfyUI from 8 GB of VRAM, with voice cloning, 16-speaker dialogue, inline cues and Qwen3-ASR.

Kokon is an open source desktop editor that runs Qwen-Image 2.1 Turbo locally through stable-diffusion.cpp: one installer, painted edits, AI upscaling and a built-in MCP server.

Fizgig v8 rebuilds the LoRA and fine-tune studio as a web app: train from a browser or your phone, borrow another computer's GPU, and open RunPod pods in the same window.

Simplified Matching Distillation is a regression-only recipe for few-step diffusion. Its students cover FLUX, Qwen-Image, Wan2.1, MiniMax H3 and 3D, with ComfyUI workflows for H3.

Alissonerdx's MiniMax H3 LoRA pack adds a v1.0 character swap that recasts a guide clip's subject plus live wallpaper adapters, with ComfyUI workflows and examples.

BFS-Character-Swap-Community-1.1

Comfy-Org's Nvidia VFX Nodes reach 0.2.1 with RTX Video Super Resolution, RTX Video HDR and RTX Video Frame Generation, an official template and Manager install.

FluxSR publishes its one-step 4x real-world super-resolution LoRA for Flux.1-dev, with training and inference code plus a ready community ComfyUI node pack.

FLUX-1-dev-non-commercial

Fish Audio releases Drama 3, a 35B MoE text-to-speech model that follows bracketed performance direction, with open weights and a ComfyUI partner node.

Fish Audio Community License 1.0

LightX2V publishes a 4-step DMD-distilled Lightning LoRA for Qwen-Image 2.1, covering text-to-image and multi-reference editing with side-by-side samples.

Apache-2.0

ModelTC releases LightVAE, compressed Wan2.1, Wan2.2 and MiniMax-H3 video VAE decoders that keep the original latent interface, up to 2.94x faster in ComfyUI.

Apache-2.0

Spark-H3 brings Spark-Attn block sparse attention to MiniMax-H3 in ComfyUI, with prebuilt nodes for RTX 40 and 50 series and a 1.73x faster DiT at 10% density.

Apache-2.0

HEISS UI is an open-source local front end for ComfyUI: a prompt box and gallery over 28 model families, with missing files one click away and no node graph to wire.

MIT

Alibaba's official Qwen-Image-2.1-Turbo checkpoint runs the 7B generator and editor in 8 denoising steps at CFG 1, with Comfy-Org repackaged weights for ComfyUI.

Qwen Research License

Akatz Labs' Person Remover LoRA tracks a person with SAM 3.1, fills the mask green and rebuilds the background in overlapping windows, with a ready ComfyUI workflow.

MiniMax H3 Community License

A community LoRA for Qwen-Image 2.1 keeps a restored or enlarged photo on its original frame: 0.3 px off against 2.2 px without it, plus a ComfyUI workflow and two checkpoints.