Spaces:
Running
is it open-source? cannot find anything?
so is it opensource? its cool, will like to use localy, for editing personal stuff.
We haven't yet released the model weights.
Our legal department is currently reviewing the privacy risks of the model and training data.
You can try it out in this space or our website first.
We offer a generous trial period for each user.
Do you guys plan to release the weights in future?
The previous version of your Space was very impressive. It modified almost everything according to my prompts. Now it modifies almost nothing except for the coloring of black-and-white photos.
Do you guys plan to release the weights in future?
Yeah, As an algorithm researcher, I hope to open source it. However, it ultimately depends on whether the legal department approves it, which requires several rounds of battles between our algorithm team and the legal department.
The previous version of your Space was very impressive. It modified almost everything according to my prompts. Now it modifies almost nothing except for the coloring of black-and-white photos.
This is strange. Our algorithm is trained to edit anything. It can change clothing, poses, colors, remove backgrounds, and so on.
We haven't updated the underlying model recently; I simply updated the Space UI to support multiple languages.
Could you show me some example prompts and images so I can test them? If it's inconvenient to post them here, my email address is [email protected]
The previous version of your Space was very impressive. It modified almost everything according to my prompts. Now it modifies almost nothing except for the coloring of black-and-white photos.
This is strange. Our algorithm is trained to edit anything. It can change clothing, poses, colors, remove backgrounds, and so on.
We haven't updated the underlying model recently; I simply updated the Space UI to support multiple languages.
Could you show me some example prompts and images so I can test them? If it's inconvenient to post them here, my email address is [email protected]
Here's a screenshot! It only partially transformed the vintage art into a photo. Before the update, it transformed it completely!
No changes here at all.
The watermark has been removed, but the digital art has not been replaced with a photo.
The watermark has been removed, but the digital art has not been replaced with a photo.
Thanks for the feedback, we will start studying these cases ！
The problem persists! It doesn't change the pose, but it removes the extra fingers!
Also does not want to improve bad photos from SD 2-1 Demo
Transformed from a photo in anime
Transformed from anime in photorealistic
The transformation of the anime into a studio photo was unsuccessful.
And that's just part of the problem!
Does not want to correct distorted photos.
She changed her pose only once! And only with her left hand!
She changed her pose only once! And only with her left hand!
Regarding pose changes: I think it's helpful to clearly specify the pose you're changing to, for example: changing the pose to sitting in a chair; changing the pose to standing sideways with hands on hips. Unlike other common image editing apps, this demo doesn't modify the user input prompt.
The watermark has been removed, but the digital art has not been replaced with a photo.
Thanks for the feedback, we will start studying these cases ！
Regarding photo style modification, especially when modifying to realistic photo styles, it seems the model has some issues understanding the instructions, which we're currently investigating.
For other issues, such as pose changes and finger repair, I think explicit prompts could improve performance. Alternatively, you can use local-inpaint to paint an area and specify the prompt for the desired modification.
It looks promising! I hope they release the model weights soon. In the meantime, check out ジオメトリーダッシュ for some fun!