Why Chat GPT image API cuts the image format after editing ?
Thanks,
Can you please share your prompts?
‘Add sunglasses’. It seems GPT image knows only 3 formats : 1024x1024
, 1536x1024
(landscape), 1024x1536
(portrait)
and the first prompt please (so the community can try to recreate)
the first image is not from Chat gpt
Hi,
I think it’s an Aspect Ratio issue
Ah, thanks for clarifying.
well yes it is. But if your input is 9:16 ? you will never get a 9:16 as the output? that’s my problem
Yes it only returns 3 sizes
that’s an interesting challenge.
any luck with adding extra directions, like “don’t crop”?
No it’s really the system that can return only 3 formats. But all other models like gemini returns your original formats. That’s stupid
This might help :~)
Create custom aspect ratio image - #4 by phyde1001.
This was for Dall-E but the same idea applies
Take a look at this topic.
The challenge is known and can be resolved using different prompting techniques.
Yeah but guys, it’s an API, not a solo usage thing. Our users will put weird prompts. You cannot rely only on prompting for displaying things inside rigid aspect ratios. For consistancy, The API must inherently support the original ratios. You are the only one to do this.
so wrap the query with additional instructions?
I got slightly luckier btw (despite the even worse aspect - which was a deliberate stress test)
Well no legs anymore lol! But thanks for trying
Actually a sensible decision imho when the aspect ratio was unhelpful.
No but I guess I will need another model than GPT, Im making an image edition tool. So you cannot change the ratio when editing an image lol
yeah, that’s not going to work at present.
but you could definitely look at techniques to prompt your way through this.
nah, It’s like you are on Photoshop, and everytime you change something on the image, It changes the ratio. makes no sense.