I am planning to make an app but I’d need masking to be available to retain the previously generated photo in order to only change small areas of it in each generation. Is masking including and working in the 4o image gen ?
Image output is not available for AI language models at the present time.
It may not be presented simply as just another output modality of a chat model where images by gpt-4o to come to the API.
The only API model that works on the edits endpoint is dall-e-2.
thanks, I just realized masking doesn’t work in 4o ? When I edit an image, mask an area and explicitly tell it to only change what’s masked, it still changes the entire image, event if slightly. Is this a known bug ?
You are talking about ChatGPT, and that is a known impossibility because that is simply not how it works.
Yeah, I’m doing this in 4o. I just tested it in Dall-e 3 and it works. But then, why is the edit / masking interface available in the 4o prompt interface ?
The image is just a visual input. The technology is significantly different than DALL-E. It could be sent into DALL-E but is not.
I had a simple request to show you, where in ChatGPT the term “mask” is not an end-user convention, and could see the part about requesting a sailboat that far exceeded any mask selection hinting - before getting shut down like many other time-wasting normal requests now.
(enjoy the original sunbathers that were unrecognizable tumors of flesh…)
1 Like