Add Stable Diffusion 3 guide for KerasHub #1955
Conversation
james77777778
requested review from
MarkDaoust,
fchollet and
pcoet
as code owners
divyashreepathihalli
left a comment
divyashreepathihalli
left a comment
There was a problem hiding this comment.
Thanks for adding this super cool guide @james77777778!! I have left a few comments.
| float16. | ||
|
|
||
| It is also worth noting that the preset "stable_diffusion_3_medium" excludes the | ||
| T5XXL text encoder, as it requires significantly more GPU memory. The performace |
There was a problem hiding this comment.
maybe also add content on if users wanted to try it with T5XXL text encoder, how can they do that.
There was a problem hiding this comment.
I've added this line:
The weights, including T5XXL, will be available on KerasHub soon.
| display_image = concate_images([np.array(image), generated_image]) | ||
| plt.axis("off") | ||
| plt.imshow(display_image) | ||
|
|
There was a problem hiding this comment.
explain the output that is seen here with 1 or 2 lines
There was a problem hiding this comment.
Added:
As you can see, a new image is generated based on the reference image and the
prompt.
| ) | ||
| plt.axis("off") | ||
| plt.imshow(display_image) | ||
|
|
There was a problem hiding this comment.
Add few lines here explaining the output.
There was a problem hiding this comment.
Added:
Fantastic! The dog is replaced by a cute black cat, but unlike image-to-image,
the background is preserved.
Note that inpainting task also includes `strength` parameter to control the
image generation, with the default value of `0.6` in Stable Diffusion 3.
|
@divyashreepathihalli EDITED: |
divyashreepathihalli
left a comment
divyashreepathihalli
left a comment
There was a problem hiding this comment.
Thanks Hongyu! LGTM
4 participants
@divyashreepathihalli
This guide covers all the APIs for SD3 in KerasHub. Please let me know if any changes are required.