GPT-IMAGE 1.5

GPT-Image 1.5 is an advanced image-text-to-image model designed to generate high-fidelity images based on both textual prompts and reference images. Developed with a strong emphasis on prompt adherence, GPT-Image 1.5 excels at producing outputs that not only closely follow the provided descriptions but also maintain the original composition, lighting, and fine-grained details present in the input images. This makes the model particularly suitable for applications where preserving visual elements from a source image while applying specific changes or creative modifications is crucial.

The model accepts two key input modalities: text and image. Text prompts allow users to describe the desired scene, action, or modifications to be made, while image URLs serve as visual references for the transformation. Users can provide one or more reference images, which the model will use to drive the image generation process. Additionally, there is support for mask images, enabling selective editing by specifying which parts of the reference image should be changed.

Users have considerable control over the image generation process through a variety of parameters. Key configuration options include image size (with support for auto, 1024x1024, 1536x1024, and 1024x1536 aspect ratios), background settings (auto, transparent, opaque), quality levels (low, medium, high), and input fidelity (low, high). The number of images generated per request can range from one to a maximum of four, letting users quickly obtain variations. Output formats supported by the model are PNG, JPEG, and WebP.

Quality characteristics are a highlight of GPT-Image 1.5. It is explicitly designed to produce high-fidelity images that respect the intent and instructions of the prompt, while ensuring that the fundamental aspects of the input image—such as composition and lighting—remain intact. Fine-grained detail preservation further sets it apart in the image-to-image domain. Users can tailor the level of quality to suit different expectations or requirements, with higher settings yielding more detailed and polished results.

Streaming output is supported, allowing users to preview results and download generated images through a simple interface. Both API and playground access are provided, enabling integration into diverse workflows or experimentation without setup overhead.

From a technical standpoint, the model's API requires at least a prompt and one image URL to function. Optional controls, such as mask images for targeted editing and sync mode for custom output delivery, provide additional flexibility. The system is configured for practical commercial and professional applications, though the documentation does not specify particular industries or user roles.

Limitations and considerations are addressed indirectly through the configuration parameters—such as a maximum of four images per request and fixed output format/image size options—which guide users in their expectations about batch processing and resolution. Best practices for prompt construction or image selection are not detailed in the documentation.

Overall, GPT-Image 1.5 delivers robust, flexible, and high-quality image-to-image generation that effectively blends visual fidelity with strong prompt alignment, supported by a straightforward yet powerful set of controls for diverse editing and creation scenarios.