Create images with GPT Image 2

GPT Image 2 is an image generation model from OpenAI, the first in their lineup built on the O-series reasoning architecture. Before rendering, the model plans the composition, looks up facts online if needed, and renders complex scenes with accurate multilingual typography. Launched in April 2026, it took first place in the public Image Arena benchmark by a record margin.

The key difference is the thinking mode and online search before rendering. The model first analyzes the prompt, plans the frame's structure, and looks up facts online if needed, and only then starts drawing. That gives industry-leading accuracy for complex compositions: comics, magazine spreads, infographics, and dense layouts with precise text.

The price depends on task complexity, resolution, and whether online search is used. On the Start and Pro plans, you get a monthly credit pack with a fixed price per image. See the pricing section for the exact cost.

Base resolution is 2K. Formats include 16:9, 9:16, 1:1, 4:3, 3:4, and more. You can generate up to eight images at once, handy for serial campaigns and quickly picking a variant.

It's a unique feature of OpenAI's O-series architecture: before rendering, the model "thinks through" the scene. It analyzes the prompt, plans the composition, and checks the logic. This takes a bit longer than a simple generation but delivers much more accurate results in complex scenes with text, infographics, and dense layouts.

If the prompt involves time-sensitive topics, real events, recognizable objects, or specific products, GPT Image 2 automatically searches the web for up-to-date information and uses it when rendering. This keeps real landmarks, locations, brands, and trends accurate.

Yes, that's one of GPT Image 2's key strengths. Text rendering accuracy is above 95%, including small fonts, curved surfaces, and dense layouts. It renders text in many languages, including Chinese, Japanese, Korean, Hindi, Arabic, Bengali, and more, and handles Latin script flawlessly.

Yes. GPT Image 2 works in image-to-image mode: upload a finished image, describe the edit, and the model carefully changes the specified elements without touching the rest. Great for changing an outfit's color, moving an object, or changing the background or hairstyle.

30 to 90 seconds, depending on task complexity. When the model needs to search the web or build a complex, multi-layered composition, generation takes longer than usual. Simple scenes with no extra checks render faster.

No. GPT Image 2 is available right inside Flami with a regular Start or Pro subscription. No need to sign up with OpenAI or pay for ChatGPT Plus or Codex separately: all models live in one account.

Every generation uses credits. On a paid plan, you get a monthly credit pack and spend it on creating images. If you run out, you can buy extra credits as a one-time purchase.

Credits refresh automatically at the start of each billing period. Unused credits from last month don't carry over, so if you didn't use up your allowance, you can switch to a smaller plan next month.

You can buy extra credits as a one-time purchase or switch to a bigger plan. The new pack becomes available right after payment, no need to wait until the end of the month.

Yes, anytime. In the Subscriptions section, turn off auto-renewal: your current paid period runs its course, and nothing gets charged after that.