Watch Weave turn AI generation from a single image into a brand content machine that one word re-runs every month
Figma deep dive: Weave | Config 2026 · Ron
41 min total·Actually worth watching closely: ~33 min·3 must-watch clips
- 0:15 – 7:23Watch
Starting on the node canvas: from chat box to workflow
Introduces where Weave sits now that it is part of Figma, opens on the off-trail hiking brand as the running case, and demos the basics of the node canvas: one prompt run into several image models in parallel to compare, then the chosen result passed to an editing model for a targeted change.
The next step for AI creation is not a better single generation, it is connecting generate, pick and edit into a workflow you can reuse
The first few minutes are spoken setup and can be sped up, but the parallel models at 229s and the hand-off to editing at 346s are the frames that make the product make sense▶ Jump to 0:15Speaker · S03 (Ron) - 7:23 – 14:29Watch
Three pieces of scale: compare, variables, iterator
Uses the compare node's slider to see the before and after of an edit, splits the prompt into a fixed part plus a changeable one exposed as an input handle, then puts a text iterator on that variable so the downstream nodes fan out in parallel, one branch per element.
The prompt variable plus the text iterator is the core mechanism that takes a creative process from one image at a time to running in batch
The iterator fanning out at 654s and the model-picking moves at 730s are the kind of operations you only grasp instantly by watching the screen▶ Jump to 7:23Speaker · S03 (Ron) - 14:29 – 21:34Watch
Hand finishing, then reusing the flow for video
Color correction, levels and crop nodes bring back fine control; the compositor opens into a Figma-like canvas for the typography, and once the input is a video the whole color and compositing chain is reused as is, with the edit happening in the same place.
A flow built for images can be handed to video unchanged, and any change upstream propagates automatically down the whole chain
1082s, where the image chain is reused for video, and 1156s, where the canvas turns into the surface for editing it, are moments no verbal summary can convey▶ Jump to 14:29Speaker · S04 (Gal) - 21:34 – 27:43Listen
Building the machine with an LLM: the brand brief as constant context
Casts the LLM as an expert in brand image analysis, turning a grid of brand images into a precise visual production brief that becomes the constant context behind every prompt the machine writes; with the Figma node wired in, the full chain runs and produces eight brand images for the new location.
The right use of an LLM in the pipeline is describing images and writing and improving prompts, not general Q&A
The value here is in the argument and most of the screen is prompt text, so you can listen rather than watch, but the first full run of the chain at 1614s is worth looking up for▶ Jump to 21:34 - 27:47 – 30:55Watch
Figma design frames become the machine's output end
The whole set of designs reused every month, the Instagram post, story, email headers and brochure, is pasted into the canvas at once; drop a generated image onto a frame and it wires to the layer. Change the image upstream and all the frames update together, and when you connect text you get the layers of that type in the design.
Design templates stop being the final layout step and become an output end of the workflow, regenerated automatically from upstream
This stretch is flagged as heavily visual: the moment at 1781s where one change updates every asset is the most convincing demo of the talk, and you have to watch it▶ Jump to 27:47 - 30:58 – 35:29Watch
One word drives it all, plus two-way editing in Figma
The input to the whole workflow narrows to a single location word that drives the images, the copy and every design frame; a weave connected frame lets you edit in Figma with your components and design system as usual and bring it back into the workflow, so a design that does not work can be fixed on the spot without touching the original frame.
The ideal shape for a complex machine is a minimal input, one word, with a human able to step into Figma and fix things at any point
1870s, one word driving the whole chain, and 1965s, jumping out to edit and back, are dense with interaction detail, and the fix at 2103s shows what the collaboration actually looks like▶ Jump to 30:58 - 35:34 – 38:59Watch
weave tools: workflows packaged as tools on the canvas
Any workflow can be packaged as a weave tool and called from the tool section right on the Figma canvas with a very simple UI; every tool, including the prebuilt ones, can be opened back into weave to read, duplicate and remix.
The tool system is open: the machine you build can become a tool others run in one click and can also take apart
The demo at 2141s, running a tool inside Figma without leaving the design environment, is the frame that shows what the product finally delivers▶ Jump to 35:34 - 39:06 – 40:31Listen
Closing: systematizing creativity, and the release timeline
Lands the point that designers have always built systems, and this is the next layer of that instinct, then gives the timeline: weave tools are in open beta, and the Figma node and publishing tools to the community roll out gradually over the next few weeks.
The Figma node and community publishing roll out gradually over the next few weeks, so you can plan when to try them
Spoken summary and release news, no need to watch the screen▶ Jump to 39:06