ComfyUI News & Open-Source AI Releases

Tencent releases Hunyuan Image 3.0, the first open-source commercial-grade native multimodal image generation model with a total of 80B parameters, featuring world knowledge reasoning capabilities and supporting complex semantic understanding of thousands of characters

Open-Weights

Nunchaku team releases optimized version of 4-Bit Lightning Qwen-Image-Edit-2509 model, supporting 4/8 step fast inference, running smoothly on 8GB VRAM environment

Alibaba Tongyi Lab launches Wan-Animate, a unified character animation framework based on Wan2.2, supporting character animation generation and video character replacement, with open-sourced model weights and inference code

Open-Weights

Comfy Cloud runs ComfyUI in your browser on hosted GPUs with pre-loaded models and custom nodes. Start free with 5 GPU runs, then pay per credit.

Tencent Hunyuan team releases Voyager technology, capable of generating world-consistent 3D point cloud sequence videos from a single image and user-defined camera paths, supporting infinite world exploration and direct 3D reconstruction

Open-Weights

ByteDance launches the USO model, capable of freely combining any subject with any style while maintaining subject consistency and achieving high-quality style transfer effects

Open-Weights

Wan2.2-S2V is an AI video generation model that can convert static images and audio into videos, supporting dialogue, singing, and performance content creation needs.

Open-Weights

Join the first ComfyUI Weekly Challenge and win $100! Create a video turning around the cat man character using the provided depth map.

The MeiGen-AI team has open-sourced the InfiniteTalk model, enabling precise lip-sync and unlimited-length video generation. It supports both image-to-video and video-to-video conversion, marking a breakthrough in digital human technology.

MIT

ComfyUI officially releases the Subgraph feature, allowing users to package complex node combinations into single reusable subgraph nodes, greatly improving workflow modularity and manageability.

Qwen-Image is a 20B-parameter MMDiT image model focused on complex text rendering and precise editing; it is now available natively in ComfyUI. This brief summarizes key capabilities, license, and resources.

Open-Weights

Tencent Hunyuan team releases open-source MixGRPO framework, the first to integrate sliding-window mixed ODE-SDE sampling for GRPO, achieving up to 71% training speedup for human preference alignment in diffusion and flow models.

Open-Weights

Black Forest Labs and Krea collaboration officially releases FLUX.1 Krea [dev] model, the best open-source FLUX model for text-to-image generation. ComfyUI has implemented native support, allowing users to experience the latest text-to-image generation technology.

FLUX-1-dev-non-commercial

WAN team officially releases Wan2.2 open source version, featuring innovative MoE architecture that brings significant quality improvements to video generation. ComfyUI has achieved native support, allowing users to directly experience the latest video generation technology.

Open-Weights

ByteDance has open-sourced the Seed-X 7B model, designed specifically for translation tasks. It supports translation between 28 languages and is suitable for integration in various scenarios such as AI image generation.

Open-Weights

PUSA V1.0 leverages innovative VTA technology to achieve high-quality, multi-task video generation with minimal data and cost, supporting image-to-video, keyframe generation, video extension, text-to-video, and more.

Open-Weights

The AMAP team releases FLUX-Text, a new diffusion-based scene text editing method supporting multilingual, style consistency, and high-fidelity text editing.

Open-Weights

OmniAvatar model is now open source, supporting audio-driven full-body virtual human video generation with natural movements and rich expressions, suitable for podcasts, interactions, dynamic scenes, and various applications.

Open-Weights

Tongyi Lab open-sources ThinkSound, the first Any2Audio unified audio generation and editing framework supporting chain reasoning, enabling multimodal inputs including video, text, and audio, with high-fidelity, strong synchronization, and interactive editing capabilities.

Open-Weights

ByteDance open-sources XVerse model, enabling precise independent control of multiple subject identities and semantic attributes (like pose, style, lighting), enhancing AI image generation's personalization and complex scene capabilities.

Open-Weights