ChatGPT Image Sizes, Aspect Ratios and Quality Settings

People search for "ChatGPT image settings" because the chat window hides most of them. There are no sliders, but there are real choices behind every image: shape, size, quality, format and background. In ChatGPT you control them in words and with a few buttons; through OpenAI's API, designers and developers can set them exactly. This guide covers both, with figures from OpenAI's help centre and developer documentation at the time of writing.

In ChatGPT: shape, background and format

• Aspect ratio: any ratio. Use the aspect ratio picker or include the ratio in your prompt ("16:9", "4:5", "1:1"). In the editor, Aspect ratio regenerates an existing image in a new shape.
• Transparent background: ask for one and ChatGPT can make the background transparent, delivered as a PNG.
• Quality and size: chosen by the service rather than by you. Paid plans (Plus, Pro, Business) can use Images with thinking for complex requests.
• Saving: every image lands in your Images library, where Save downloads it to your device.

Choose the ratio from where the image will be used: 1:1 for shop tiles and profile images, 4:5 for social feeds, 9:16 for stories and reels, 16:9 for slides and video thumbnails, 3:2 for editorial images. Setting it before generating composes the picture for that frame; cropping afterwards throws composition away.

Through the API: every setting

OpenAI's current API image models are GPT Image 2.5 in two versions: Sunburst, recommended where editing precision matters most, and Flare, for fast, high-quality everyday generation. The output settings:

OpenAI marks resolutions above 2560x1440 as experimental, and notes that xhigh and max quality are new with GPT Image 2.5; earlier GPT Image models go up to high. Size, quality and background accept auto, in which case the model chooses based on the prompt.

Quality against cost and time

OpenAI's advice is to use low quality for quick drafts and to compare higher settings for final assets, balancing detail, latency and cost. In the API, cost is token-based and varies with model, quality and size, so the same prompt can cost very different amounts. Complex prompts can take up to two minutes. A sensible routine: draft at low quality and a recommended size, approve the composition, then render once at high or above at the final size.

Sunburst or Flare

For API work the choice of model matters as much as the size. OpenAI positions Sunburst for workflows where editing precision matters most, such as repeated changes to an approved product image, and Flare for fast, high-quality everyday generation, such as many variations of a social post. Both share the same token rates, but token consumption differs by model and quality, so test both on a typical request before committing a batch.

Transparent backgrounds that work

1. In ChatGPT, say "transparent background" in the prompt and describe the subject as a cut-out: "a single ceramic mug, isolated".
2. Through the API, set the background to transparent and the format to PNG or WebP; JPEG cannot carry transparency.
3. Check the edges on a dark and a light surface. Glass, hair and fine details are where cut-outs fail.
4. Tidy the edge in a design tool if needed, and keep the transparent master for reuse.

Product mockups that use transparency are covered in product mockups in ChatGPT.

When the image is not big enough

ChatGPT's native output suits screens and social media. For print, an outside upscaler is the usual answer: work out the pixels the print needs (inches × 300 for close viewing), choose a faithful upscaling mode for approved images, and check text at 100% afterwards. Through the API you can request larger custom sizes first, up to the limits above, which reduces how much upscaling is needed. Our comparison of AI image upscalers covers the options.

Common size problems

• The image came out square. The ratio was not set; use the picker or put the ratio at the start of the prompt.
• Text is soft at the edges of a wide image. Very wide or tall ratios spread detail thin; generate closer to 3:2 and extend in the editor, or set text in a design tool.
• A transparent PNG has a faint halo. Fine edges such as hair and glass are hard to cut out; place the image on its final background before approving it.
• A custom API size is rejected. Check that both sides are multiples of 16, the ratio is within 1:3 to 3:1 and the total pixel count is within OpenAI's limits.

Record the settings

An image is only reproducible if you know how it was made. With each delivered file, note the prompt, the ratio, whether thinking was on, and for API work the model, size, quality, format and background. ChatGPT's Images library keeps images only as long as their conversations exist, so keep your own record too.

Try it

A single matte terracotta plant pot, isolated, soft even studio light, subtle shadow beneath, transparent background, product cut-out style.

Generate it in ChatGPT with a 1:1 ratio, download the PNG and place it on a dark and a light background to check the edge. Then ask for the same pot at 4:5 with the aspect ratio picker and compare how the framing changes.