GPT Image 1.5

OpenAI's most capable image generation model. Up to 4x faster than gpt-image-1 with 20% lower API costs. Optimized for image editing with precise modifications while maintaining consistency in lighting, composition, and facial features. Handles denser and smaller text rendering. Can output both text and images.

This was the release that made editing the point of an image model rather than a side feature. OpenAI's headline claim was precise edits that keep everything else intact: ask for a change and the model changes that, holding lighting, composition and the appearance of people steady across the original, the output and every subsequent edit. Generation also got up to four times faster than the model it replaced.

The lab lists the edit types it targeted - adding, subtracting, combining, blending and transposing - and the uses that follow from doing them reliably: practical photo retouching, believable clothing and hairstyle try-ons, and stylistic or conceptual transformations that keep whatever made the original worth editing. It also reported more reliable instruction following than the first ChatGPT Images model, which shows up less in single objects than in compositions where the relationships between elements have to survive, and another step forward in text rendering, specifically denser and smaller text.

Through the API the same model is aimed at repeatable commercial work. OpenAI describes it as stronger at image preservation and editing than the previous generation, with more consistent preservation of branded logos and key visuals across a sequence of edits - which is the whole problem in marketing and brand work, and in ecommerce teams generating a full catalogue of variants, scenes and angles from a single source image. Image inputs and outputs were priced 20% below the previous model. Both text and images can come back from a single call, which its successor does not do.

OpenAI was unusually candid about what it had not fixed. It reran the examples from its first image-generation launch and published the limitations alongside the improvements, naming style reproduction, scenes with multiple faces, and multilingual output as cases that still fell short, and said flatly that results remain imperfect with significant room for improvement. Its current guidance is to keep GPT Image 1.5 for workflows already validated against it and to prefer the newer model for new work, particularly where editing reliability or flexible output sizes matter.