OpenAI GPT Image 2.5 Flare Image to Image: AI Image Editor
Edit images with GPT Image 2.5 Flare. Fast, precision single-element edits that keep the rest of the scene untouched. Coming soon on Muapi — no subscription.
Edit images with GPT Image 2.5 Flare. Fast, precision single-element edits that keep the rest of the scene untouched. Coming soon on Muapi — no subscription.
About this model
GPT Image 2.5 Flare Image to Image is OpenAI's newest instruction-based image editing model, announced September 8, 2026. It changes only what you ask for while keeping the rest of a scene untouched, and keeps earlier edits intact across a multi-turn chain instead of drifting or degrading like GPT Image 2. GPT-Image-2.5 Flare is OpenAI's default, fastest tier — the same quality and editing gains as GPT Image 2.5 at up to 50% lower latency than GPT Image 2. Muapi is bringing this endpoint online as soon as OpenAI opens general API access. See GPT Image 2.5 Sunburst for the higher-precision tier.
Cost analysis
| Provider | Cost | Notes |
|---|---|---|
| muapiapp | TBD at launch | Coming soon — general availability pricing not yet published by OpenAI. |
| Fal.ai | Not available | GPT Image 2.5 is not yet available on Fal.ai. |
| Replicate | Not available | GPT Image 2.5 is not yet available on Replicate. |
Coming soon — general availability pricing not yet published by OpenAI.
GPT Image 2.5 is not yet available on Fal.ai.
GPT Image 2.5 is not yet available on Replicate.
** Competitor pricing is estimated based on similar model architectures and usage tiers.
Configuration schema
| Parameter | Type | Description | Default |
|---|---|---|---|
| Prompt | string | Text instructions describing the desired transformation. Maximum 20,000 characters. | Change only the beverage bottle in the center to a matte black can labeled "PEAK", keeping the exact same lighting, shadows, reflective surface, background props, and camera angle. Do not change anything else in the scene. |
| Image URLs | array | Upload or provide input images to transform. Up to 16 images supported. | https://d3adwkbyhxyrtq.cloudfront.net/webassets/videomodels/gpt-image-2-image-to-image-in.jpg |
| Aspect Ratio | Enum (6 options) | Output image aspect ratio. Note: images with aspect ratio 'auto' (or unspecified) will only be converted to 1K; 1:1 cannot be converted to 4K — otherwise the task will fail to create. | auto |
| Resolution | Enum (3 options) | Image resolution. Note: images with a 1:1 aspect ratio cannot be converted to 4K. Images with aspect ratio 'auto' (or unspecified) will only be converted to 1K; otherwise the task will fail to create. | 2K |
| Quality | Enum (3 options) | Generation quality. 'low' and 'medium' use a faster, cheaper backend; 'high' uses the higher-fidelity backend. | high |
Text instructions describing the desired transformation. Maximum 20,000 characters.
Change only the beverage bottle in the center to a matte black can labeled "PEAK", keeping the exact same lighting, shadows, reflective surface, background props, and camera angle. Do not change anything else in the scene.Upload or provide input images to transform. Up to 16 images supported.
https://d3adwkbyhxyrtq.cloudfront.net/webassets/videomodels/gpt-image-2-image-to-image-in.jpgOutput image aspect ratio. Note: images with aspect ratio 'auto' (or unspecified) will only be converted to 1K; 1:1 cannot be converted to 4K — otherwise the task will fail to create.
autoImage resolution. Note: images with a 1:1 aspect ratio cannot be converted to 4K. Images with aspect ratio 'auto' (or unspecified) will only be converted to 1K; otherwise the task will fail to create.
2KGeneration quality. 'low' and 'medium' use a faster, cheaper backend; 'high' uses the higher-fidelity backend.
highDeveloper documentation
Provide a source image — upload or link up to 16 reference images.
Write a precise instruction — state what should change and what must stay identical (e.g. "replace only the bottle, keep the lighting and background exactly as shown").
Submit the request — once live, call the endpoint below and poll for the result:
curl -X POST https://api.muapi.ai/api/v1/gpt-image-2.5-flare-image-to-image \
-H "x-api-key: YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{"prompt": "Change only the jacket color to forest green, keep everything else identical", "images_list": ["https://example.com/source.jpg"]}'
request_id against /api/v1/predictions/{id}/result until status is completed.Frequently asked
It is OpenAI's newest instruction-based image editing model, announced September 8, 2026 alongside ChatGPT Images 2.5, available via the GPT-Image-2.5 Flare and Sunburst API models.
Not yet. This endpoint is in early access with OpenAI. Muapi will enable it as soon as general API access opens — no action needed on your part.
GPT Image 2.5 is more reliable at changing only the requested element while preserving the rest of a complex scene, and keeps earlier edits intact across multiple turns instead of drifting.
Flare is the default, fastest tier with the same quality and editing gains as GPT Image 2.5. Use [GPT Image 2.5 Sunburst](/playground/gpt-image-2.5-sunburst-image-to-image) if you need the higher-precision tier instead.
Yes. [GPT Image 2 Image to Image](/playground/gpt-image-2-image-to-image) is live now while GPT Image 2.5 is in development.
Create a free Muapi account and API key today — the gpt-image-2.5-flare-image-to-image endpoint will activate automatically for existing users once OpenAI opens general availability.