ComfyUI shipped four point releases between November 10 and November 15 — v0.45.0 through v0.45.4 — and nearly every line in the changelog is the same shape: a new commercial model, wired in as a Partner Node, callable from inside the same node graph artists already use for local, open-weight workflows. Seedream 5.5, GPT Image V3, MiniMax H3 Image, LTX’s AlphaGen, and now Claude Fable 5.2, HeyGen Video 1.0, and three new Meta Muse models have all landed in roughly a week, per ComfyUI’s own changelog.
What actually shipped
The headline additions in v0.45.4 (November 15) are a new node for Claude Fable 5.2 on the Anthropic Claude node, a general-purpose video model from HeyGen that the company describes as its first model built for broad use rather than its signature talking-head avatars, and three Meta Muse variants — Wonder, Spark 1.3, and Ember — added to the OpenRouter LLM node. Earlier in the week, v0.45.0 and v0.45.1 added ByteDance’s Seedream 5.5, OpenAI’s GPT Image V3 (plus new GIF and SVG output nodes), and LTX’s AlphaGen VFX suite with alpha-channel output. Comfy also moved Krea’s and Wan’s node families onto its newer “Router” backend, which the changelog frames as infrastructure work rather than a user-facing feature.
Why the pace is the real story
No single one of these additions is dramatic on its own — a new text-to-video model, a new LLM option, a backend migration. What’s notable is the cadence: four releases in six days, each adding partner-model coverage rather than new local capability, on a project whose core appeal has always been that it’s free, open-source, and runs on a GPU you own. Partner Nodes don’t change that — they’re opt-in, billed through Comfy Credits, and ComfyUI itself stays free either way — but the sheer breadth of what’s now reachable from inside one node graph (Anthropic’s chat models, Google’s Gemini, ByteDance’s and MiniMax’s generative media, HeyGen’s talking-head and now general video, Meta’s open LLMs) means a ComfyUI workflow can now orchestrate most of the commercial AI-media market without leaving the canvas.
What this means for a working artist
For a technical artist already living in ComfyUI, the practical upshot is fewer context switches: a workflow that needs a large-language-model step, an image pass, and a video pass can increasingly stay in one tool instead of stitching together three separate web apps and their separate billing. That consolidation is quietly becoming ComfyUI’s second identity — not just a local diffusion front end, but a general orchestration layer for AI media, open and closed alike.