OpenAI brings ChatGPT's image generator to developers

When OpenAI quietly rolled out its latest image generator inside ChatGPT last month, users went wild. Over 130 million people generated more than 700 million images in the very first week.

Now, OpenAI is unlocking the engine behind that viral feature for everyone else. The company officially launched gpt-image-1 across its API, while Microsoft simultaneously rolled it out to gated enterprise customers on Azure AI Foundry.

This is not just another minor DALL-E checkpoint. It is OpenAI's natively multimodal image generation model, built specifically to take on commercial design workloads directly inside third-party apps.

Real text rendering and multi-turn editing

If you have used DALL-E in the past, you know the pain: gibberish text whenever you wanted a simple sign, and virtually no way to feed an existing image back into the model for iterative editing.

gpt-image-1 fixes both fronts. The new model brings reliable text rendering directly into generated graphics, making it genuinely useful for posters, book illustrations, and UI mockups. According to Microsoft's technical specs, it supports minimum resolutions of 1024 pixels, including square 1024×1024, portrait 1024×1536, and landscape 1536×1024.

More importantly, developers get true image-to-image workflows. You can supply an existing image alongside a text prompt to edit it, transform styles, or run inpainting with user-drawn bounding boxes.

Big software platforms are already jumping in. Canva is tapping gpt-image-1 to turn rough user sketches into polished graphic assets inside Magic Studio. GoDaddy is building automated logo design and typography tools for small businesses, while HubSpot, Instacart, and invideo are using it for marketing collateral and video assets.

Pricing is structured by tokens rather than a flat per-image fee. Prompt text costs $5 per 1 million tokens, input image tokens cost $10 per million, and generated image tokens cost $40 per million. In practice, OpenAI says that works out to around $0.02 for a low-quality square image, $0.07 for medium quality, and roughly $0.19 for high quality. Images also ship with C2PA metadata baked in by default.

Why it matters

For developers and digital product builders, gpt-image-1 fills a massive gap. Until now, building an app that required accurate in-image typography and iterative image edits usually meant wrestling with self-hosted Stable Diffusion pipelines or dealing with brittle workarounds. Having an official API that handles both zero-shot prompt precision and image editing lowers that barrier overnight. That said, at nearly twenty cents for a top-tier render, running it at high consumer scale will chew through margins fast.

For the broader AI landscape, this puts immediate heat on competitors. Midjourney has famously resisted opening a standard developer API, while Google's Imagen has mostly lived inside its own cloud silo. OpenAI is pairing conversational intelligence with graphic production right where enterprise developers already build.

We think OpenAI's decision to price this natively via tokens is smart, because it lays the groundwork for seamless interleaved multimodal workflows where text and image tokens flow together. However, with open-weights models like Flux offering stunning visual fidelity at a fraction of raw compute cost, OpenAI will need to prove that its convenience and enterprise safety controls are worth the premium.