Yep, same here. The feature is completely useless with a soft mask. Who came up with that?
Even their example is pointless, you can just say “put a flamingo in the pool” and it will give the exact same result
1 Like
I think it’s because the underlying tech is different. DALLE was pure image model, and 4o is multi-modal… Pretty sure it is a soft mask. I’m trying to get more details, so stay tuned…
You don’t have to be pretty sure, one api call and you are 100% sure. The whole image gets regenerated
2 Likes
Yeah, I’m not sure if the masked area is weighted or what, though. Like I said, trying to get an OpenAI employee to give us some more details if possible.
I wonder if they would consider adding an edit endpoint for DALLE3 in the meantime… I’m still using that model for quite a bit personally.
1 Like
since it’s calling the same underlying image endpoint as ChatGPT, lessons learned from there apply (you’re just providing your own prompt, which ChatGPT does for you, but the tool invocation has been identical for me).
The mask doesn’t do anything, and in my experience, NOT providing a mask, and instead telling the generator what to change, and to change nothing else, will frequently result in better outcomes than even providing a mask. It’s not built like inpainting at all.
I know this is “API Forum” but this particular tool is different. Both ChatGPT and Image-1 appear to call the same underlying remote tool, so anything we learn on one is applicable to the other:
2 Likes
I’m echoing others in this thread. I’m also experiencing issues with the Image Generation API when providing a mask. It seems the mask is being ignored and the entire image is being edited.
3 Likes
Let’s be honest: this isn’t a bug — it’s false advertising.
OpenAI’s API does not currently support true inpainting, despite what the documentation claims.
According to the docs, only the transparent areas of the mask should be modified, while the filled areas remain unchanged. But that’s clearly not what happens.
The official examples use small images where the issue is easy to miss — but if you view them at full resolution, the problem becomes obvious.
Even in the official example, the entire image is regenerated — not just the masked (transparent) area — directly contradicting the claim that “the filled areas will be left unchanged.”
3 Likes
Any update from openai on if this will be fixed? Its pretty unfortunate and has wasted a few hours of my time trying to get it to work.
there is probably no solution. I tried to edit an image with selection feature on chatgpt app and i got a fully re-rendered image st the end.
for reference @klaroy other image models can do this such as flux
I have the same problem (for the record)
@wendyjiao can we get confirmation if this is a bug or expected behaviour so we can plan accordingly to work around this please?
Hi, thank you everyone for reporting the issue and participating in the discussion.
We are looking into this issue right now. We will shortly update this thread on whether this is expected behaviour or whether there the API is not being called correctly.
Thank you!
Dear Community,
We have tested the reported issue and would like to confirm that in order to achieve the desired edits, as of today, the workaround would be to use the dall-e-2 model.
For example, for images provides in this message the result would look like this:
You would need to ensure the images are compatible with Dalle-2 (one image, and it should be a square png file less than 4MB).
We are validating with Engineering whether this is also available or could be made available with gpt-image-1 and will update this thread shortly. We hope that the provided workaround unblocks some customers and allows for experimentation.
Thank you for your patience and understanding.
1 Like
Appreciate the response!
Just FYI dall-e-2 is not a viable workaround for me due to the significant difference in the quality of generations, I assume the same is true for many others.
1 Like
We’d like to provide an update to everyone on this thread: the behavior observed with gpt-image-1 has unfortunately been confirmed as a known limitation. While the functionality performs as expected when using dall-e-2 - albeit with some compromise in image quality - the issue described here is indeed present with gpt-image-1. We sincerely apologize for the inconvenience this may have caused.
3 Likes
Then what should and must be done is to restore dall-e-2 to working order. It has been significantly damaged in the last few months from the quality of a year and a half ago, before the release of dall-e-3.
A post you can look at with a prompt vs image 2 years ago:
Then:
What is obtained today:
And infill of a “polar bear” now infills a blob of nonsense. Openai-image inpainting example is not working - #2 by _j
Try to remake these “voice avatars” created with dall-e-2?"
A attractive woman of South Asian heritage poses for her photograph in a studio with a mottled grey background. The woman has dark straight hair, and wears an Indian wedding dress.
4 Likes
Thanks for clarifying @OpenAI_Support
That’s quite disappointing. As you can see from this thread and similar ones on this forum, a lot of people rely on this masking functionality to work with gpt-image-1. Are there any plans to address this?
1 Like
We have plans to implement precise in-painting, however, cannot commit to any timelines right now. Please stay tuned for updates!.
9 Likes
Love to hear that.
Thank you guys for the support!