ComfyUI News & Open-Source AI Releases

Sand AI has open-sourced MAGI-1, an autoregressive model that generates videos chunk by chunk, offering 24B and 4.5B parameter versions and supporting multiple video generation modes

Open-Weights

SkyworkAI releases new open-source video generation model SkyReels-V2, supporting infinite-length videos with both text-to-video and image-to-video capabilities

Open-Weights

Nari Labs introduces open-source text-to-speech model Dia 1.6B, capable of generating multi-character dialogues directly from text with emotion and non-verbal communication

Open-Weights

Kunlun Wanwei's SkyReels team releases and open-sources SkyReels-V2, the world's first infinite-length film generative model using Diffusion Forcing framework, capable of producing high-quality, long-duration video content

Open-Weights

Shakker Labs launches the new FLUX.1-dev-ControlNet-Union-Pro-2.0 model with optimized control effects, support for multiple control modes, and smaller model size

Open-Weights

InstantCharacter is an innovative tuning-free method that enables character-consistent generation from a single image, supporting various downstream tasks

Open-Weights

Developed by Lvmin Zhang, FramePack technology compresses input frame context, making video generation workload invariant to video length, allowing processing of numerous frames even on laptop GPUs

Open-Weights

Seaweed-7B achieves performance surpassing 14B parameter models with only 7 billion parameters, at just one-third the training cost of industry standards, bringing new possibilities to the video generation field

Open-Weights

A new video inpainting framework FloED has released its code and weights, achieving higher video coherence and computational efficiency through optical flow guidance

Open-Weights

PixelFlow innovatively operates in raw pixel space, simplifying the image generation process without requiring pre-trained variational autoencoders, enabling end-to-end trainable models

Open-Weights

HoloPart can decompose 3D models into complete, semantically meaningful parts, solving editing challenges in 3D content creation

Open-Weights

UniRig uses autoregressive models to generate high-quality skeleton structures and skinning weights for diverse 3D models, greatly simplifying the animation workflow

Open-Weights

OmniSVG is a new unified multimodal SVG generation model capable of producing highly complex, editable vector graphics from various inputs including text, images, or character references

Open-Weights

Researchers develop TTT-Video model using Test-Time Training technology based on CogVideoX 5B, capable of generating coherent videos up to 63 seconds long

MIT

ByteDance Creative Intelligence team releases UNO model, unlocking greater controllability through in-context generation, achieving high-quality image generation from single to multiple subjects

Open-Weights

Tiamat AI team releases EasyControl framework, adding conditional control capabilities to DiT models, now supported in ComfyUI via the ComfyUI-easycontrol plugin

Open-Weights

HiDream.ai releases HiDream-I1, a new open-source text-to-image model with 17B parameters that outperforms existing open-source models in multiple benchmarks, supporting high-quality image generation in various styles

Open-Weights

Stable-X team introduces Hi3DGen, an innovative framework for generating high-fidelity 3D models from images, addressing the lack of geometric details in existing methods through normal bridging technology

Open-Weights

TripoSF, based on the innovative SparseFlex representation, supports 3D model generation at resolutions up to 1024³, capable of handling open surfaces and complex internal structures, significantly improving 3D asset quality

Open-Weights

Kunlun Wanwei releases the world's first commercial-grade controllable video generation framework SkyReels-A2, enabling multi-element video generation through dual-branch architecture, bringing new possibilities for e-commerce, film production and more

Open-Weights