GPT Image 2.5
GPT Image 2.5 is OpenAI's precision-focused image model. It is built for premium visual work, extra fidelity on intricate detail, and edits that stay scoped to the instruction while preserving subject and composition across many rounds of revision.
Overview
| Property | Value |
|---|---|
| Provider | OpenAI |
| Cost | 40 (low), 200 (medium), 400 (high), 800 (extra high), 1200 (max) |
| Modality | Image |
| Prompt Required | Yes |
| Modes | Text-to-image + image editing |
| Output | Blue Image |
Endpoint Routing
Armox automatically switches endpoint at runtime:
- No input image ->
openai/gpt-image-2.5/sunburst/text-to-image - With input image(s) ->
openai/gpt-image-2.5/sunburst/edit
You do not need to pick a separate model for edit mode.
What It's Best For
- Premium stills with fine typography and intricate detail
- Instruction-scoped edits that keep the original subject and composition
- Multi-round revisions with reference images
- Transparent-background cutouts
Inputs
Required Inputs
| Input | Type | Description |
|---|---|---|
| Prompt | Yellow Text | Describe what to generate or how to edit the image |
Optional Inputs
| Input | Type | Description |
|---|---|---|
| Input Images | Blue Image | Up to 16 reference images. Attaching any image switches to edit mode |
Configuration
| Setting | Type | Default | Description |
|---|---|---|---|
| Image Size | select | auto | Use Auto in edit mode to match the input image |
| Quality | select | medium | Higher settings increase detail, latency, and credit cost |
| Background | select | auto | Auto, opaque, or transparent |
| Output Format | select | png | jpeg, png, or webp |
Quality
| Option | Credits |
|---|---|
| Low | 40 |
| Medium | 200 |
| High | 400 |
| Extra High | 800 |
| Max | 1200 |
| Auto | 200 |
Medium is 200 credits and High is 400. High is the default in Canvas and the Edit App. Extra High and Max cost more because they use more image tokens.
Background
Use Transparent with PNG output when you need a cutout. In the Edit App, background removal already requests a transparent background from this model.
Output
The model outputs a Blue Image that can connect to:
- Other image models
- Image tools such as upscalers
- Video models that accept an image input
Tips & Best Practices
- Keep quality at High unless you want a cheaper or more expensive tier
- Attach a clean source plus an annotated guide when editing in the Edit App or the Inpaint tool — GPT Image 2.5 follows the instruction instead of a binary inpainting mask
- Prefer PNG when requesting a transparent background