GPT-4o Image Generation: What It Was and What Replaced It
GPT-4o image generation launched in ChatGPT in March 2025. What it did well, how OpenAI's GPT Image models replaced it, and how to get similar results today.
The Short Answer
GPT-4o image generation was the image feature OpenAI built directly into GPT-4o. It launched in ChatGPT on March 25, 2025, and took over from DALL·E as ChatGPT's default image generator. On April 23, 2025, OpenAI brought the same natively multimodal model to its API as gpt-image-1.
OpenAI has released several image models since, and its own 4o announcement now points readers to ChatGPT Images 2.5 for the current experience. Spicy Studio doesn't offer GPT-4o image generation. It offers the two newest models in that line, GPT Image 2.5 and GPT Image 2, and they're how you get that kind of result today.
What GPT-4o Image Generation Did Well
OpenAI's pitch was images that are useful, not only pretty:
Legible text. Street signs, menus and invitations with real, readable words.
Prompt following. OpenAI said GPT-4o could handle 10 to 20 different objects in one image, where other systems struggled with around 5 to 8.
Editing by conversation. Because generation lived inside the chat model, you could refine a picture over several turns and keep a character consistent.
Working from uploads. It could transform an image you uploaded or use it as visual inspiration.
OpenAI also listed limits, among them posters cropped too tightly, hallucinated details, weaker non-Latin text, imprecise edits and dense small text. Renders were slow, often up to a minute (OpenAI's announcement).
From GPT-4o to GPT Image 2.5: The Timeline
March 25, 2025: 4o image generation arrives in ChatGPT.
April 23, 2025: gpt-image-1 brings that model to the API.
October 6, 2025: gpt-image-1-mini adds a cheaper API option.
December 16, 2025: a new version of ChatGPT Images, with GPT Image 1.5 in the API.
April 21, 2026: ChatGPT Images 2.0 and gpt-image-2, with stronger dense and non-Latin text.
September 8, 2026: ChatGPT Images 2.5, offered in the API as Flare for fast everyday work and Sunburst for precise edits.
As of October 2026, OpenAI's model list shows GPT Image 2.5 Sunburst, GPT Image 2.5 Flare and GPT Image 2 as current and marks GPT Image 1, 1 Mini and 1.5 as deprecated. Dates are from OpenAI's announcements and API changelog.
Getting GPT-4o-Style Results on Spicy Studio
Both successors are in the Image Lab, paid with credits, with no OpenAI account or API key.
GPT Image 2.5: start here. Flare is the default version; switch to Sunburst for precise edits. You get 1K, 2K or 4K, quality from low to max, and up to 16 references at 1 credit each. An image costs 4 credits at 1K low, 9 at 1K medium and 56 at 2K high, rising to 350 at 4K max.
GPT Image 2: 1K or 2K, low to high quality, 5 to 60 credits, and up to 16 references at no extra charge, which makes it cheaper for edits that combine many uploads.
New accounts get 25 free credits with no card, enough for a few 1K drafts. The GPT Image 2.5 guide and the GPT Image 2 guide cover the settings in depth.
Prompt Tips: Text and Multi-Reference Edits
Quote the exact words and describe the lettering. “A hand-painted wooden sign outside a bakery that reads ‘FRESH SOURDOUGH DAILY’ in cream serif capitals, early morning street, shallow depth of field.” Short copy is easier to get right than a paragraph, and finals with text belong at high quality.
Proofread everything. OpenAI's documentation says the models can still struggle with precise text placement and clarity.
Edit with references instead of chat turns. In the Image Lab you upload images and describe the change. Name each upload by position: “Use the perfume bottle from the first image and the logo from the second. Place the bottle on a marble shelf in the room from the third image, soft window light.”
Say what can't change. “Keep the bottle's shape, label and color exactly as in the first image.” Then change one thing per pass, and use Sunburst for the last, most precise edit.
That three-reference edit costs 26 credits on GPT Image 2.5 at 2K medium (23 plus 3 for the references) and 25 on GPT Image 2, where references are free.
Frequently Asked Questions
Can you still use GPT-4o image generation?
+
Not as it launched. OpenAI has upgraded ChatGPT's image feature several times since March 2025, most recently to Images 2.5, and its 4o announcement now points there. In the API, gpt-image-1, the model that followed 4o image generation, is marked deprecated as of October 2026.
What replaced GPT-4o image generation in the API?
+
gpt-image-1, released on April 23, 2025. As of October 2026, OpenAI's current API image models are GPT Image 2.5 Flare, GPT Image 2.5 Sunburst and GPT Image 2.
Does Spicy Studio offer GPT-4o image generation?
+
No. Its Image Lab offers OpenAI's current models, GPT Image 2.5 (Flare and Sunburst) and GPT Image 2, plus other image models such as Nano Banana 2.1.
Which model should I use instead of GPT-4o for text in images?
+
Start with GPT Image 2.5 on Flare and render text you'll publish at high quality. Pick GPT Image 2 when an edit uses many references, since it doesn't charge for them.
Want to run this yourself? Every new account starts with free credits.
Open the Image Lab