AI moves into the tools you already use.

Generation comes to the timeline

Using a generator has usually meant leaving your editor, generating something on a website, downloading it and dragging it back in. Monday's launches cut out that round trip. Krea's Premiere Pro plugin lives in Window → Extensions and is pitched at the gaps in an edit: cutaways, inserts, title sequences, background plates for composites, and trial looks before anything is committed to the timeline. Krea lists similar plugins for Photoshop, After Effects, Illustrator, DaVinci Resolve and Blender on the same page.

Figma took the other route: instead of adding a panel, it opened its Weave models to agents. Per the Weave setup guide, an agent connected through the Figma MCP can call Seedance, Kling, Nano Banana and the rest, but it needs a paid Weave account, spends Weave credits, and asks you to confirm the cost before it runs. That confirmation step is worth noticing as agents start spending credits on your behalf.

Apps you describe instead of download

Runway's App Studio goes a step further: you describe a tool, and Runway writes, tests and hosts it on its own models, Brand Kits and Workflows. The built-in examples (Character Swap, AutoComp, Product Portfolio, 3D Camera Control, Precise Edit) read like a list of the single-purpose apps people were paying for separately this year. It comes with paid plans and uses existing credits.

ComfyUI's answer is to give the recipe away. OPENjutsu is a clonable character-swap app and development guide, MIT-licensed, built on Akatz Labs' H3 Character Swap workflow. It is also a starting point for Comfy's #OpenSourceHiggsfield challenge, with entries due 30 October. The catch for hobbyists is hardware: it was tested on a 96 GB GPU and deploys through the Comfy Developer Platform. A community batching node that stretches H3 motion transfer to 447 frames shows where the local crowd is pushing.

MOSS-Image becomes a toolkit

The MOSS-Image community is now building controls around the model rather than just posting samples. Moss Film Looks turns film stocks and lenses into presets, LayoutPrompt gives each element its own box, and a QwenVL node writes the layout JSON from an existing picture. ComfyUI v0.43.0 added rope scaling and colour-depth control for the model. The most interesting idea may be Toska, which has an agent paint stroke by stroke in Krita instead of generating pixels in one go.

Taste, and who gets credit

The most-shared design posts were about not looking generic. A thread arguing that AI-built websites all look the same passed 100,000 views, and a Claude Code design-studio stack drew more bookmarks than likes. The pixel-art sprite clean-up recipe was saved more than 1,200 times for the same reason: the raw output isn't the finished asset. Outside the tools, the arguments were about money and credit: an AI-generated feature in China flopped at the box office, A$AP Rocky's video was accused of reusing another artist's footage, and Elvis Presley's rights holders sued Anthropic over training.

Tell us what to dig into next: reply with a medium, a tool or a question and we'll steer tomorrow's sweep toward it.