OpenAI is rolling out a major overhaul to its image generation capabilities today, launching a new flagship model, dubbed GPT Image 1.
OpenAI is rolling out a major overhaul to its image generation capabilities today, launching a new flagship model, dubbed GPT Image 1.5, that promises to fix the most frustrating limitations of previous versions: speed, consistency, and text rendering.
The new ChatGPT Images experience is available to all users starting today, bringing generation speeds up to four times faster than before. But the real competitive edge lies in precision.
The core promise of GPT Image 1.5 is that it can finally handle complex, multi-step edits without destroying the original image’s integrity. When users upload a photo and ask for changes, say, swapping a shirt or adding a hat, the model now adheres to the intent reliably, preserving crucial elements like lighting, composition, and even facial likeness across subsequent edits. This consistency is the holy grail for practical AI image use, enabling believable clothing try-ons and professional-grade photo edits that don't require starting from scratch every time.
Perhaps the most immediately useful improvement for professional users is the massive leap in instruction following and text rendering. AI image generators have historically struggled with complex layouts, grids, and, most famously, legible text.
GPT Image 1.5 demonstrates a new ability to handle dense, smaller text and complex formatting, even rendering detailed markdown into a believable newspaper article layout. This moves the tool beyond simple artistic creation and into the realm of graphic design and marketing mockups, where accurate text placement is non-negotiable. The source material shows the model successfully executing a complex 6x6 grid of specific objects, a task that would have been impossible for earlier models.
For users, OpenAI is streamlining the workflow by introducing a dedicated "Images" creation space within ChatGPT. This new sidebar feature includes preset filters and trending prompts, making creative exploration less reliant on perfect written prompts.
The improvements extend directly to the enterprise market. The API version, GPT Image 1.5, is now 20% cheaper for image inputs and outputs. Companies like Wix are already leveraging the model’s improved consistency for brand work, noting that the ability to preserve logos and key visuals across edits is essential for generating product catalogs and marketing materials at scale.
While OpenAI acknowledges that results remain "imperfect" in areas like scientific accuracy and rendering multiple small faces, this release represents a significant acceleration in the generative AI arms race. By tackling speed, consistency, and text, the three major workflow bottlenecks, the new ChatGPT Images makes a strong case for being the most capable general-purpose image model currently available.
GPT‑Image‑1.5 is OpenAI’s latest image generation model, available in ChatGPT and via API for creating and editing images from natural language prompts. It focuses on higher visual fidelity, better prompt adherence, and faster generation compared with the prior GPT‑Image models.
The 1.5 release improves instruction following, text rendering inside images, and detailed control over edits like lighting, composition, and facial consistency. It also runs up to around 4x faster than the previous model in many typical use cases
You can generate new images from scratch, create concept art, design mockups, product visuals, and stylized illustrations using text prompts. You can also iteratively edit existing images to add or remove elements, adjust style, or refine composition while preserving the original structure.
GPT‑Image‑1.5 is exposed in ChatGPT through a dedicated Images entry point with preset styles and templates. Developers can call it through the API using the GPT‑Image‑1.5 model name in image generation and editing endpoints.
Yes. You can take an existing image and ask for precise edits such as “make the lighting cooler,” “change the hairstyle,” or “add a logo in the corner,” while keeping other aspects consistent. The model is optimized to maintain identity, lighting, and composition across multiple edit rounds
GPT‑Image‑1.5 significantly improves legible text generation in images, including titles, labels, and infographics. It can handle smaller fonts and denser layouts more reliably than the earlier version, though occasional spelling or layout errors can still occur.
The model targets enterprise scenarios like marketing creatives, product visualization, and design iteration with more consistent, controllable outputs. Improved precision editing and instruction adherence make it better suited for workflows that require brand consistency and repeatable visual styles.
Yes. The model enforces safety rules around violence, adult content, hate imagery, and misuse of real people, which may cause some prompts to be refused. Images may also include provenance or watermarking mechanisms as the ecosystem evolves, to help indicate AI origin in downstream use
GPT‑Image‑1.5 competes with models like Google’s Nano Banana Pro and other leading generators, aiming for faster iteration and more reliable edits. Benchmarks and early testing suggest it closes gaps in speed and control, especially for iterative creative workflows.
Get the most important AI news daily.
