Start with the saved image, not a screenshot
Use the image file you saved from ChatGPT after the final edit. Choose a JPEG, PNG or WebP above. A screenshot is a separate image and can reduce resolution; Transfa can clean the supported original export without recompressing it.
The file details show which removable containers were found, the before/after size and the unchanged dimensions. “No removable metadata” is a valid result, not a statement that the photo contains no AI content.
What about OpenAI’s provenance signals?
OpenAI documents Content Credentials and invisible watermarks for supported generated images, with coverage depending on the product, model and export path. Transfa removes supported attached metadata containers. It does not remove a watermark embedded in the image or verify provenance signatures.
Reference: OpenAI’s provenance documentation ↗. Reviewed October 17, 2026. File-cleaning support is based on Transfa’s implementation; platform-label outcomes are separate.
Use the new copy for a new upload
Save the file ending in -clean, then choose that file when posting. Re-exporting it through another editor may attach new information. For Instagram, see the posting and phone-saving steps. A successful cleanup does not guarantee a label-free post.
Before you save
Does this remove a ChatGPT watermark?
No. It cleans supported metadata containers. Invisible watermarks and visible image content are unchanged.
Which ChatGPT image model does it support?
The cleaner checks JPEG, PNG and WebP file structure, not the model version. Export format support is not a certification of every ChatGPT model or export path.