Model guide
GPT Image: candidates, transparency and delivery
Distinguish quality and file controls with packaging candidates and transparent PNG/WebP assets.
Open featureView current models and settings in the workspaceOn this page
GPT Image: instructions, edits, and delivery formats
| Model | Tasks to try | Settings and limits here |
|---|---|---|
| GPT Image 2 | Text-driven creation, multi-reference combinations, edits, and multiple candidates. | Low/medium/high quality, exact dimensions, and candidate count. PNG output; no transparency, format, or compression selector. |
| GPT Image 2.5 Flare | General creation, multi-reference edits, transparent materials, and delivery formats. | Low/medium/high/extra-high/max quality, exact dimensions, candidates, PNG/JPEG/WebP, background, and JPEG/WebP compression. |
| GPT Image 2.5 Sunburst | Another model to compare for the same creation or editing goal. | The same currently exposed settings as Flare. Equal controls do not imply equal results or a better model. |
Usage, cautions and actual examples
These models currently have no web-search, manual-thinking, or separate negative-prompt controls. Instruction understanding does not imply a toggle for those features.
For 2.5 transparency choose the actual background option with PNG/WebP. A prompt saying 'transparent' or a white picture is not alpha; inspect the file.
Candidates are alternatives, not sequential narrative. Check width, height, ratio, and total pixels together. Compression controls delivery bytes rather than model quality.
For transparent materials, define subject and edge requirements, choose PNG/WebP and Transparent, then inspect preview and file alpha before raising quality. Recheck background if later switching to JPEG.