GPT Image 2.5 multi-round edits: chain edits on one image via API
To make several edits to one picture with GPT Image 2.5, feed each result back as a reference. What fal says Sunburst preserves, and what to check each round.
To make several edits to one picture with the GPT Image 2.5 API, chain the calls: send the original with your first instruction, then send the returned image URL as the reference for the second instruction, and so on. Each call is independent, so keep every instruction small and name what must stay the same.
fal's page for Sunburst edit describes the model as built for edits "scoped precisely to the instruction, with subject and composition preserved across many rounds of revision". That is fal's description, not a Sume measurement. Request fields come from Sume's Image API docs. Read 2026-09-29.
How do I feed a result back in?
A finished 200 response returns data[].url, a Sume-hosted signed URL. Put that URL in input_references for the next round. Signed URLs can expire, so run a chain promptly or download each result and re-host it at a public HTTPS URL you control.
curl -X POST "https://api.sume.com/v1/images" \
-H "Authorization: Bearer $SUME_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "openai/gpt-image-2.5-sunburst",
"prompt": "Now make the mug matte black. Keep everything else identical.",
"input_references": [
{ "type": "image_url", "image_url": { "url": "https://example.com/round-1-result.png" } }
],
"aspect_ratio": "auto"
}'What should I check after each round?
- Subject: is the face, product or logo the same as in the original?
- Composition: did the framing or camera angle move?
- Detail: fine text and small patterns are the first things that drift.
- Keep the round-1 original. If round 4 drifts, restart from the original with a combined instruction.
| Check | Why |
|---|---|
| Subject identity | Face, product shape and logo are what edits tend to move |
| Framing | Composition can shift between rounds |
| Fine text | Small lettering is redrawn each time |
| Original kept | Restart from it if drift builds up |
Is Sunburst required for chained edits?
No. Both ids accept the same request. OpenAI describes Sunburst as its most capable model for image generation and editing, and Flare as its model for everyday generation. Run the same chain on both ids and keep the one whose drift you can live with.
Does each round cost the same?
Each round is a separate generation billed on its own tokens, including the reference image you send back in. A failed or cancelled generation is not billed. See the price breakdown for output figures.
Sources
Related posts
More in Developers
- GPT Image 2.5: how many images can I generate per request?
On Sume, GPT Image 2.5 takes n from 1 to 4 images per request, billed per image. How to ask for several versions and what a 202 means for a larger n.
- GPT Image 2.5 API: OpenAI's two endpoints versus Sume's one route
OpenAI serves GPT Image 2.5 on images/generations and images/edits. On Sume both become POST /v1/images. How the fields map and what returns instead of base64.
- GPT Image 2.5 output format: png, jpeg or webp on Sume
GPT Image 2.5 on Sume returns png, jpeg or webp through output_format. Which to pick for transparency and speed, and what output_compression does on Sume today.
- GPT Image 2.5 Python example: generate an image with the Sume API
A runnable Python example for GPT Image 2.5 on Sume: async httpx, POST /v1/images, read data[].url, and stop cleanly on a 202 job. Copy, set the key, run.
Written by Sume