PROMPT LIBRARY

GPT Image 2.6 prompts, each with the image it made.

Each prompt is quoted word for word from the model maker's guide or a creator's post, next to the output it produced. Open a card for the full prompt.

OpenAI released GPT Image 2.6 on November 4, 2026. In ChatGPT it is called Images 2.6 and adds animated GIFs and reference images. The API has two models with the same price: gpt-image-2.6-flare for speed, as fast as six seconds an image, and gpt-image-2.6-sunburst for more detail and steadier edits. Both work on the generations and edits endpoints and add markup edits, where you circle the part to change, SVG export, GIF output and a complexity setting. Price per million tokens: text input $5, image input $8, image output $30. The prompts below are OpenAI's own guide examples, shown with the Sunburst output, and animated GIFs creators shared on X, each linked to its post.

Sources: OpenAI on X · OpenAI on X (API) · API: gpt-image-2.6-sunburst · API: gpt-image-2.6-flare · Image prompting guide

We measured the GIFs on this page frame by frame: frames, loops and length →

93 prompts

How to write prompts for this model

Bring your GPT Image 2.5 prompts over unchanged

OpenAI's 2.6 guide keeps all 24 example prompts from its 2.5 guide word for word and adds three, for circled edits, looping GIFs and complexity. Its migration advice is to rerun your existing prompts with the same reference images, size, quality and output format, try 2.6 Flare where you used 2.5 Flare and 2.6 Sunburst where you used 2.5 Sunburst, and change one setting at a time. Source: AgentCraft: GPT Image 2.5 vs 2.6 prompts · OpenAI: Migrate an existing workflow

Name the result first, then what it looks like

OpenAI asks you to name the subject and the intended use, such as a product photograph, an advertisement or a diagram, then the composition, aspect ratio and placement, then materials, lighting, colors and the visual medium. For complex requests it suggests labeled sections: scene, subject, details and constraints. Source: OpenAI: Prompting fundamentals

Pick a length and format you can maintain

OpenAI says short prompts, paragraphs, JSON-like structures and tags can all express the same intent, so choose the one that is easiest to read and update. For scale: its own examples have a median of 71 words over 4 lines, and the 36 creator prompts for still images on this page have a median of 274 characters. Source: OpenAI: Prompting fundamentals · AgentCraft: GPT Image 2.5 vs 2.6 prompt data

Quote the exact text and say how often it appears

Put the required wording in quotes, describe its position and typography, spell unusual words or brand names letter by letter, and ask for no extra text. OpenAI suggests comparing medium or high quality when the text is small or dense. On this page, all 8 GIF prompts that asked for on-image text got it legible, in Chinese, Japanese and English. Source: OpenAI: Render exact text · AgentCraft: GIF prompts measured

For edits, say what changes and list what stays

Write change only X, then list what to preserve: identity, geometry, layout, lighting or labels. With several uploads, number each image and give it a role, such as subject, style, clothing or background. 12 of OpenAI's 26 examples here and 20 of the 67 creator prompts start from an uploaded image. Source: OpenAI: Prompting fundamentals · AgentCraft: GPT Image 2.5 vs 2.6 prompt data

Circle the area instead of describing where it is

New in 2.6: draw a circle around the object on a transparent overlay, send it as markup with the photo, and describe only the change. OpenAI's example recolors a striped vase to cobalt blue and asks to keep its shape, the stems and the rest of the kitchen, then to remove the red circle. Source: OpenAI: Guide an edit with markup

For a GIF, write one line first

31 of the 67 creator prompts on this page make a GIF, and their median length is 68 characters. 大阪城を建築する流れをGIFにして (build Osaka Castle as a GIF) is a whole prompt and came back as a 12-frame sequence. Add a frame grid or describe the loop only when the first result misses; the questions below cover both. Source: AgentCraft: GPT Image 2.5 vs 2.6 prompt data · AgentCraft: GIF prompts measured

Refine one change per turn, then check

Pass the previous output as the next input, ask for one change, and repeat the details to preserve; restate critical constraints if the result drifts. Before you use the image, check that every word reads as requested, labels and relationships are right, identities and product shapes survived, and the edit changed only what you asked. Source: OpenAI: Refine an image across turns · OpenAI: Check the result

Questions about this model

What is GPT Image 2.6?

OpenAI's image model released on November 4, 2026. In ChatGPT it is called Images 2.6. In the API it comes as two models, gpt-image-2.6-flare and gpt-image-2.6-sunburst, which both generate and edit images from text and image inputs. Source: OpenAI on X

Should I use Flare or Sunburst?

OpenAI describes Flare as the model for fast iteration and Sunburst for detailed visual work. Both support complexity control, markup-guided edits, SVG output and animated GIF output, on the generations and edits endpoints, at the same price. Source: OpenAI image generation guide

How much does GPT Image 2.6 cost in the API?

Per million tokens, for both Flare and Sunburst: text input $5 ($1.25 cached), image input $8 ($2 cached), image output $30. The complexity setting has no separate fee. Source: API: gpt-image-2.6-sunburst

How do I get an animated GIF from the API?

Set output_format to gif on the generations or edits endpoint, and describe the subject, the motion and the loop in the prompt. frame_count takes 2 to 16 frames and defaults to 9; frame_duration_ms takes 20 to 2000 milliseconds and defaults to 200. To guide the animation with still references, send them to the edits endpoint. Source: OpenAI image generation guide: Generate animated GIF output

How many frames does a ChatGPT Images 2.6 GIF have?

Of the 43 GIFs on this page, 22 have exactly 9 distinct frames and 36 have 4, 6, 9, 12 or 16, the cell counts of a square or near-square grid. Median length is 1.4 seconds. Name the grid in the prompt, such as 6 frames in 3 columns and 2 rows, when timing matters. Source: AgentCraft: GIF prompts measured frame by frame

How do I make a GIF loop without a jump?

Describe the last frame as the first one, for example: first and last frame match so it loops seamlessly. Among the GIFs we measured, 2 of 15 that asked for a loop jumped visibly at the restart, against 10 of 28 that did not ask. Source: AgentCraft: GIF prompts measured frame by frame

What does the complexity setting do?

It controls how much the model plans before rendering: low, medium or high, and high if you leave it out. OpenAI suggests lower complexity for focused edits such as changing a background or a sign, and higher for dense text, complex layouts and relationships between many objects. Source: OpenAI image generation guide: Set image complexity