GoKawiilFigma trains custom AI model to edit designs directly on canvas
Figma is developing an in-house editing agent trained via supervised fine-tuning to understand its native file format and make iterative design changes, such as spacing, color, or layout edits, directly on the canvas. The company built training data by pairing starting designs with synthetic edit requests and validated outputs, using larger teacher models, deterministic validators, and visual judges to check results. Figma reports that an early Llama 3.3 70B model trained with this approach performed well on these tasks.
GoKawiil's interpretation of the reporting above, not reported fact.
Figma's approach suggests that general-purpose frontier models struggled to natively understand Figma's file syntax, pushing the company toward specialized fine-tuning rather than relying on costly prompting. This could let Figma embed faster, more reliable AI editing directly into designers' daily workflows, potentially reducing friction in iterative tasks that are hard to verify automatically. It may also signal a broader industry pattern of companies building narrow, domain-specific AI models rather than depending solely on general-purpose ones.
- Figma built a custom AI agent specifically trained to understand and edit its native file format.
- The company used supervised fine-tuning with synthetic edit requests rather than relying on general-purpose model prompting.
- An early version based on Llama 3.3 70B reportedly achieved strong results using this training method.
Source: figma.com, 2026-11-01
Published there as: “We're Training a Figma-Fluent Editing Model”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.