Just caught up with the April ComfyUI updates. Dropping that many models at once is a bit hard to digest, lol.

Just went through the April updates for ComfyUI wrapped before May Day—holy crap, the amount of new stuff this month is insane. Jotted down a few that I might actually use.

On the video side, Seedance 2.0 came in as a partner node on day one, with three modes: text-to-video, reference-to-video, and first-last-frame-to-video. Happy Horse and Wan 2.7 also got partner nodes. For images, Baidu’s Ernie-Image dropped full weights—it’s an 8B DiT, so if you like tinkering, you can fine-tune it. OpenAI’s GPT Image 2 handles text-to-image and editing in one flow. Then there’s Quiver, which directly outputs SVG vectors—great for icons and charts without worrying about blur when zooming in. As a half-assed programmer, that’s pretty handy.

Music-wise, Ace Step 1.5 XL is open-source but aimed at commercial quality. For restoration, SUPIR is there to salvage blurry images when paired with an upscaler. RIFE and FILM now have native frame interpolation support, and SAM 3/3.1 segmentation is in too. The most practical feature is that the API can run multiple workflows in parallel now—batch users who test parameters are probably gonna cry tears of joy. There are nearly 500 workflows on the Hub now, so whatever you want to do, someone’s probably already built it. If you actually tried to test them all, you’d need to take a week off.

Been waiting for parallel execution forever. Used to queue up variants and they’d run all the way till sunrise.

Quiver can straight up generate vector icons now, I’m loving this. No more raster upscaling turning everything into a pixelated mess.

Yeah, same here. I’ve been having some digestion issues too.

SUPIR paired with an upscaler is seriously legit. The results I got from restoring old photos turned out way better than I expected.