Formalwear, same child
Describe the finished composition in your brief: the main subject, light source, setting and intended visual style.
ReferenceEditedTurn a written brief into a visual direction with GPT Image 2.5. Explore poster layouts, product scenes and photo edits, with image settings available before you submit.
These examples are starting points, not promised outcomes. Bring your own subject and references, describe the change clearly and review the result before using it.

Input
ResultStarting from the Flare poster, change the coral glass sphere to cobalt blue while keeping the typography and lighting.
Change only the coral glass sphere to cobalt blue glass. Keep the headline, footer, paper sculpture, lighting, camera, composition, and every other detail unchanged.
These examples are starting points, not promised outcomes. Bring your own subject and references, describe the change clearly and review the result before using it.
Describe the finished composition in your brief: the main subject, light source, setting and intended visual style.

ReferenceEditedUse reference photos for edits and specify the details to keep. Review the result for changes outside your requested area.

ReferenceEditedChoose output settings to suit the asset. When requesting a transparent background, download the original PNG and inspect its edges.

ReferenceEditedPlan the subject, lighting and composition before generating. Change one instruction at a time so you can compare versions and keep the details that matter.
Text to image
Image to image
Image to image
Text to image
Image to image
Image to image
Text to image
Text to image
Text to image
Text to image
Text to image
Text to image
Text to image
Image to image
Text to imageDescribe the finished composition in your brief: the main subject, light source, setting and intended visual style.
Use reference photos for edits and specify the details to keep. Review the result for changes outside your requested area.
Try simple product or portrait scenes before adding more visual instructions. This makes revisions easier to compare.
Both models share Nano Banana 21 workspace features—text and reference photo editing, 5 fixed standard aspect ratios, and transparent PNG exports—with distinct optimizations for turnaround speed versus dense detail.
Test in generator above| Workflow & Capability | GPT Image 2.5 Flare | GPT Image 2.5 Sunburst |
|---|---|---|
| Primary Creative Focus | ✓ | High-precision typography, micro-labels & intricate art |
| Turnaround & Generation Latency | ✓ | Extended generation time for dense detail passes |
| Native Transparent Background (PNG) | ✓ | Supported in PNG |
| Reference Photo Editing (Up to 4 images) | ✓ | Supported with high detail adherence |
| Standard Aspect Ratios (1:1, 4:3, 3:4, 16:9, 9:16) | ✓ | Same 5 fixed standard aspect ratios |
| Upfront Credit Estimation & Pricing | ✓ | Credits are estimated before generation based on the mode, quality, aspect ratio, and number of reference images, with applicable membership discounts. Each run generates one image. |
These examples are starting points, not promised outcomes. Bring your own subject and references, describe the change clearly and review the result before using it.
Start from a blank canvas with a descriptive prompt up to 1,000 characters, or upload up to 4 reference images to enter edit mode.
Choose opaque or transparent background, and set quality (low, medium, or high) to balance turnaround and detail level.
Choose 1:1, 4:3, 3:4, 16:9, or 9:16. Each run generates one image.
Your request is dispatched through the asynchronous fal queue with real-time status tracking.
Download the generated PNG, or upload it again for another edit.
Plan the subject, lighting and composition before generating. Change one instruction at a time so you can compare versions and keep the details that matter.
Turn a written brief into a visual direction with GPT Image 2.5. Explore poster layouts, product scenes and photo edits, with image settings available before you submit.
Plan the subject, lighting and composition before generating. Change one instruction at a time so you can compare versions and keep the details that matter.
These examples are starting points, not promised outcomes. Bring your own subject and references, describe the change clearly and review the result before using it.
In edit mode, you can upload up to 4 reference images with natural-language instructions up to 1,000 characters. The model modifies the requested elements—such as attire, lighting, or specific objects—while preserving the core subject and composition without requiring manual mask painting.
Nano Banana 21 supports 5 fixed standard production aspect ratios: square 1:1 (1024×1024), landscape 4:3 (1024×768) and 16:9 (2560×1440), and portrait 3:4 (768×1024) and 9:16 (1440×2560), across low, medium, and high quality presets. Custom pixel inputs and auto-dimensions are not supported.
Commercial use is subject to OpenAI content policy, Nano Banana 21 terms of service, your rights in any uploaded inputs, and applicable laws. Users may utilize their outputs for creative and commercial projects, but the platform cannot provide copyright guarantees.
Flare delivers lower latency for rapid creative ideation, concept exploration, and responsive photo editing. Sunburst allocates more inference compute to fine details, rendering dense typography, intricate labels, and complex compositions.
Each run generates one PNG image. Choose low, medium, or high quality; PNG also supports transparent backgrounds.
Experience fast Flare inference, precision Sunburst typography, and reference editing directly in the workspace.