Image generation
Agents create raster images — mockups, illustrations, icons, social graphics — as part of the work they deliver.
When a task calls for a visual asset, your agents make it on the spot from a text description — no separate design tool, no round trip to a human. Photos, illustrations, concept art, icons, and social graphics all come out as image files saved alongside the rest of the deliverable.
What you can ask for
Agents describe what they want — style, composition, colors, content — and get back a finished image. You control the shape and finish:
- Orientation — square (1024x1024), landscape (1536x1024), or portrait (1024x1536), or let the model choose.
- Quality — high for detailed final art, medium for a balanced result, low for fast drafts.
- Transparent background — a PNG with an alpha channel, useful for logos and icons that need to sit on any background.
Models
Several image models are available, and the agent picks the right one for the job:
- gpt-image-2 (default) — general-purpose photos and illustrations.
- qwen-image-2.0 — strong at rendering readable text inside an image.
- qwen-image-2.0-pro — highest quality, with native 2K output.
Images are meant for genuine visual assets. For text-heavy pieces — reports, structured documents, data diagrams — agents write files and produce a preview instead, so the output stays editable.
Who has it
Image generation is turned on automatically for Designer, Developer, and Queen Bee agents, so any hive with one of those roles can produce visuals without extra setup.
Where it fits
Most useful inside content and design playbooks: a blog post that ships with its own header, a landing page with hero art, a social campaign with graphics. See how agents assemble the rest of the deliverable in documents and reports, and which roles carry the skill in hives.