Image Compression and Output Resolution

Resolution is determined by the model name

4All API image resolution is selected via the model suffix; do not rely on the size parameter:

Suffix	Description
None	Standard resolution (about 1K)
-2k	2K high definition
-4k	4K ultra-high definition (GPT-Image series)

Use this in combination with the aspect-ratio suffix, such as gpt-image-2-portrait-2k and gemini-3.1-flash-image-square-2k. For the full rules, see Image and Video Generation Models.

How to choose a tier

• Web/social media display: the standard tier is enough; it loads fast and costs less;
• Printing / enlarged crops / main product images for e-commerce: start at 2K;
• Finished assets that need post-processing: 4K, but both unit cost and processing time increase significantly;
• It is recommended to iterate on prompts with the standard tier until satisfactory, then generate the final result in high resolution—avoid repeated trial and error at 4K.

Output format and size

• When a URL is returned, download the original image directly; when base64 is returned, note that the response body is large (4K can reach several MB), so the client and proxy body size limits should be increased;
• After obtaining the original image, compress and generate multiple specs on your own (webp/avif + thumbnails) before uploading to the CDN; do not send the original image directly to the frontend;
• Generated image URLs are time-limited, so be sure to transfer them to your own object storage first.

Common issues

• Blurry images: most likely the standard tier is being upscaled on the frontend; switch to the 2K/4K tier;
• Gray edges on white-background images: specify in the prompt “pure white background #FFFFFF, no shadows”;
• File size too large: converting to webp on the server side (quality 80) can usually reduce size by 60–80% with no visible loss.