GPT Image 1.5 is OpenAI's image generation and editing model. It can create new visuals from text and make targeted modifications to existing images. It is especially suited for tasks that need to preserve people's appearances, product features, brand marks, and original compositions, while improving complex instruction following and dense text rendering to help marketing, e-commerce, and design teams continuously refine the same set of visual assets.
Submit requests to the public API at api.acedata.cloud using the documented parameters, then use the results in your application.
Input parameters and result formats vary by service. Use the public API for this model and follow its guide for generation, task retrieval and editing operations.
Image creation and detail preservation are model capabilities; the number of reference images, file formats, and result formats are invocation specifications for this platform's endpoint.
GPT Image 1.5 excels at adding, removing, combining, or replacing elements in existing images while preserving lighting, composition, and people's appearances as much as possible. When changing clothing, backgrounds, or product colors, clearly specify what must remain unchanged so the edit focuses on the target rather than redesigning the entire image.
For creative requirements involving multiple objects, clear positional relationships, or segmented layouts, the model places greater emphasis on relationships between elements. Describe subjects, spatial positions, visual hierarchy, and style separately for advertising concept images, composite visuals, and informational posters; complex tasks should still be checked for accurate object counts and placement.
The model improves rendering of smaller, denser text and enhances preservation of brand marks and key visuals during editing. It is suitable for developing headlines, product images, and layouts together, reducing drafts where text and visuals feel disconnected. Before formal publication, proofread all text item by item and check whether mark details have changed.
Provide a product image and a description of the target scene, specifying that the shape, color scheme, and branding must be preserved while changing only the environment, props, or display angle. This can be used to create candidate images for product catalogs, holiday scenes, and presentation assets in different styles for team selection; product details should be compared item by item with the physical product to avoid treating generated variations as actual specifications.
Provide the campaign theme, headline copy, brand images, and layout requirements, and have the model generate advertising visuals containing text or modify the background and decorative elements of an existing poster. The deliverables can serve as social media assets, campaign visual drafts, and design proposals, followed by review for text accuracy, information hierarchy, and brand consistency.
Based on a portrait photo, describe changes to clothing, hairstyle, environment, or artistic style, and clearly require that facial features and pose be preserved. Suitable for creating styling proposals, character visual drafts, and candidate creative portraits. When making successive edits, keep the original image for comparison to help identify deviations in appearance or local details.
If the task focuses on precise edits to existing images, preserving brand elements, or visual creation containing substantial text, GPT Image 1.5 is more worthwhile to test first than GPT Image 1. Its improvements focus on instruction following, image preservation, and editing quality; if the existing workflow already meets your needs, compare the results using the same set of assets before deciding whether to make adjustments.
When there are no visual assets yet and you want to explore the overall direction, choose text-to-image generation; when the product, person, or composition has already been determined and only local content needs to be changed, choose image editing. For detail-sensitive tasks, try the high input fidelity setting and write the changes and elements to preserve separately, avoiding the assumption that high fidelity means pixel-level invariance.
Provide a prompt when generating; provide both image and editing instructions when editing. Clearly specify the original text, subject preservation requirements, and target aspect ratio.
Use the image generation or image editing endpoint, explicitly specifying model=gpt-image-1.5; use auto or WIDTHxHEIGHT for size, and generate one image first before evaluating. Configure masks, quality, and file format according to the guide for this endpoint.
Read the URL or Base64 image according to the response format, save the task_id asynchronously before querying the result; check text, reference details, and the alpha channel, and record usage according to the current Pricing rules.
Use the original image as the basis, preserving the packaging, logo, perspective, and all text; change only the tabletop background to off-white, and add soft side lighting while keeping the shadows natural.
Choose input_fidelity=high in supported editing requests, but still compare the logo, small packaging text, and outlines item by item; fidelity control is not pixel-perfect locking.
Submit model=gpt-image-1.5 and prompt to /openai/images/generations. The prompt should describe the subject, background, lighting, layout, and any text that needs to appear; when there are no constraints from an original image, this method is better suited for exploring the complete image and visual direction.
Submit image and prompt to /openai/images/edits, and explicitly specify model=gpt-image-1.5. image can use a single image URL or an array of URLs; the instructions should separately specify the elements to change and the parts that must be preserved, such as people, products, and composition.
The editing endpoint allows up to 16 image URLs. You can provide references for people, products, or environments, and explain in the prompt the role of each image, which elements need to be combined, and which cannot be changed. When there are many reference images, avoid conflicting style and appearance requirements.
You can try creating posters with Chinese titles and copy. The model has improved dense and small-font text rendering, but multilingual performance still has limitations. Clearly provide the original text, text positions, and hierarchy, and check every character after generation; long body text or strict brand fonts should be typeset separately.
You can choose url to obtain the image address, or choose b64_json to obtain encoded image data, and select PNG, JPEG, or WebP according to your use case. Responses may contain created/data or may return task_id; applications should handle completed images and task responses separately, and should not treat a task ID as an image result.
Choose a development approach or application suited to this model and follow its setup guide.
↗Python SDK Integration Guide
SDK installation, authentication and examples↗TypeScript SDK Integration Guide
SDK installation, authentication and examples
↗↗Pricing
$0.0523· per generation
You pay$13,584for 142,800 credits · $0.0951 / credits46% OFF
You're billed in credits. Pick a top-up tier to see its per-credit rate — the larger the top-up, the cheaper each credit.
This is only a basic call example. See the full docs for more parameters and advanced usage.
View Docs