Prompt-to-vector generation in Illustrator: a text prompt becomes editable vector scenes, subjects and icons, arranged in logical groups. I worked on the segmentation behind that grouping, from the 2023 beta to general availability in 2025.
General availability
- 24 Apr 2025, Adobe MAX London
- Illustrator 29.5 (April 2025 release)
- Adobe Firefly Vector Model
- Scenes, subjects and icons
Adobe made Text-to-Vector generally available on 24 April 2025 at Adobe MAX London, and the April 2025 release of Illustrator (29.5) lists Text to Vector Graphic as powered by the latest Firefly Vector Model. It turns a text prompt into editable vector scenes, subjects and icons, arranged in logical groups so each result can be edited further.
On the way from beta to GA, Illustrator 28.6 (July 2024) added Style Reference, a detail slider, colour and tone controls and a generate-similar option. Later releases let users choose between Firefly Vector 3 and Firefly Vector 4 (Illustrator 30.0) and brought partner models into the feature.
What I did
- I built and shipped the segmentation model that splits each generated scene into clean, editable groups.
- With one colleague, I owned the model end to end: architecture, training, fine-tuning, integration, and performance and stability.
- I'm a named inventor on related Adobe patent applications on real-time image segmentation and on segmenting vector images by searching their group hierarchy.
Patents
- Generating segmentations of semantically relevant objects from vector images using vector hierarchy searching
- Real time image segmentation
Beta
- Adobe MAX, 10 Oct 2023
- Illustrator 28.0 (October 2023 release)
- Adobe Firefly Vector Model
- Scenes, subjects, icons and patterns
Adobe announced Text to Vector Graphic (beta) and the Adobe Firefly Vector Model at Adobe MAX on 10 October 2023, and shipped the feature in Illustrator 28.0. It turns a text prompt into editable vector art: scenes, subjects, icons and, at launch, tileable patterns. Adobe presented the Firefly Vector Model as the first generative AI model focused on vector graphics.
Adobe described the output as editable, compact and organised into groups, with related paths grouped together, new content placed on its own layer and gradients kept as editable vector gradients. A Generative Match option made new art follow the style of the existing design. The beta was open to Creative Cloud members whose plan or trial included Illustrator.
What I did
- I built the grouping and segmentation step that turns a generated image into clean, grouped, layered vectors.
- I fine-tuned and combined several segmentation models and wrote the logic that resolves overlapping masks, so each generated image breaks into sensible groups.
- I'm a named inventor on an Adobe patent application for hierarchical semantic grouping in image vectorization.
Patents
Sources
- Adobe revolutionizes AI-assisted creativity, Apr 2025 (Adobe Newsroom) ↗
- Generate scenes, subjects and icons (Adobe Help) ↗
- Illustrator release notes (Adobe Help) ↗
- New creative frontiers in Illustrator, Jul 2024 (Adobe Blog) ↗
- Adobe releases next generation of Firefly models, Oct 2023 (Adobe Newsroom) ↗
- Try out the latest Illustrator features, Oct 2023 (Adobe Blog) ↗