For image / pure-image / image-generator / image-action modes, skip the
task-planning round trip and hand the prompt straight to the image model,
deriving the requested image count from the composer settings. Also
normalize completed image-generator nodes back into regular image nodes
(clearing the image-generator layer role and defaulting Tone to "visual")
when merging generated results and when restoring finished nodes, so a
finished generator node is not re-appended as a duplicate.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Derive an ordered map of each inline reference image to the surrounding
prompt text describing its role, and feed it into the decision, research,
task-plan, conversation and image-prompt builders. The image prompt now
carries this placement so reference roles are not swapped.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Server: forward canvas/reference images to the image model so extract,
continue, and edit requests stay faithful to the source product.
- Add ImageRequestOptions.Images and route requests with inputs to the
/v1/images/edits endpoint, via multipart upload (default) or JSON
image_url payload, selectable through Agent.Image.InputImageTransport.
- Extract reference URLs from prompt directives, inline @image tokens,
and chat message image/mask contents; fold them into the idempotency
key and prepend a reference-fidelity instruction to the prompt.
Frontend:
- Shrink CanvasToast to a compact top pill.
- Paste an internal transparent clipboard node instead of re-uploading
when it matches the clipboard image.
- Keep agent-project polling alive on stream error while streaming.
root