ComfyUI v0.36 adds loops, Yue2 music and Gemini 3.8 hooks

ComfyUI has quietly become one of the default orchestration layers for serious image and now video generation workflows, and v0.36.0 is a meaningful expansion of what builders can do without writing custom glue code. Generic loops, richer video primitives and support for new models like Yue2 and Marigold v2 make it easier to prototype complex, multi step pipelines that mix vision, depth and audio. The partner nodes for Gemini 3.8 Flash and Microsoft’s MAI image models are also a tell: the open tooling ecosystem is racing to make proprietary frontier models pluggable alongside open ones.

This matters for the AGI race because orchestration environments shape what kinds of systems get built on top of raw models. As more creators and researchers standardize on tools like ComfyUI, the barrier to experimenting with multi modal agents that see, plan and act falls further. While this release is not a core capability breakthrough, it lowers friction for large numbers of developers to chain models, tools and media together into more agentic systems. Over time, the availability of mature open tooling around both open and closed models can narrow the gap between what well resourced labs and independent teams can build, especially in areas like generative video and interactive media.