Just scrolling through the monthly updates and holy crap, ComfyUI integrated a ton of stuff in May — video, image, 3D, audio, multimodal, you name it. Let me quickly round up the ones I actually care about.
Krea 2 is Krea’s first base model, and it’s all about “how it looks” rather than “what it drew” — that whole style reference and mood board thing. Netflix open-sourced VOID, which is pretty interesting — it’s for removing objects from video, not just erasing pixels but also cleaning up the shadows and reflections that come with the subject. Apache 2.0. Tripo 3.1 is a three-in-one: text/image/multi-view to 3D model. Luma’s UNI-1 isn’t a diffusion model — it’s a decoder-only autoregressive transformer, supposedly way better at editing and reference handling.
The rest: Claude’s now in ComfyUI so you can do prompts and reasoning in your workflow, OpenRouter has one node connecting to 20+ big models, Google’s Gemma 4, HiDream, Stable Audio 3, BiRefNet for matting, and Microsoft’s MoGe for single-image 3D geometry. ComfyHub already has over 500 workflows. At this update pace, feels like your workflow isn’t even finished before the models get swapped out again.