Replies: 8 comments 12 replies
-
|
I'm looking forward to it too, but it's still in draft mode as far as I can see. I don't know when it will be integrated into the master build, but I'd like to try it out. |
Beta Was this translation helpful? Give feedback.
-
|
Yes, it isn't still officially released but I was so hyped that couldn't wait. |
Beta Was this translation helpful? Give feedback.
-
|
Would you mind showing us some examples of cases where that feature achieves good results vs previous results, please ? I understand your hype but did not have time to experiment yet. |
Beta Was this translation helpful? Give feedback.
-
|
For special effects Old method doesn't mantain the coherence
New method, mantains the coherence
For posing characters in ilustration Left original. Pormpt Blonde woman sit, both pictures with the same seed. Same denoising strength and CFG. New method preserves the pose
Left original. Pormpt Photography of Superman, both pictures with the same seed. Same denoising strength and CFG. New method preserves the pose
A dream come true for many people combining the 2 methods, amazing for product desing. New method mantain many traits of the original.
Similar idea using Loop back. Original method One of the biggest frustrations with img2img is that the final picture doesn't retain many characteristics from the original at high values of denoising strength so if we want to mantain these characteristics we need to worlk with low values, which affect to the quality of the picture,. With this method many characteristics are preserved even with high values of denoising strength, and we can mix both methods (new and old) to find a balance in the final result. I find this new method specially interesting in posing characters for illustration. Because it tries to mantain the original pose without many strange artifacts. I would like to see if it can work with animation mantainning the coherence between frames. It can also work very well to change the artistic style of a picture mantainnig the original traits of the picture. And also for replacing characters in a picture without the need of masks. |
Beta Was this translation helpful? Give feedback.
-
|
Great work thank you very much. I'm unsure about what you mean by "combining the 2 models". |
Beta Was this translation helpful? Give feedback.
-
|
I was refering to the new and the old method. I don't know how it will be in its final release. In its actual state you have in settings an option called "Blend betweeen an unmasked and masked conditioning image for inpainting models.", it goes from 0-the new mehtod to 1-the old method. In the new method the traits are preserved but sometimes it can be very strict with respect the original picture, and not create the changes that we want. The old method is less strict but creates a lot of modifications in the final picture. A combination of both can be useful, for example if we want to transform a 2D picture in something more 3D. Or if we want big changes in the final picture trying to mantain characteristics of the original. |
Beta Was this translation helpful? Give feedback.
-
|
Ok thank you now it makes sense |
Beta Was this translation helpful? Give feedback.
-
|
I can't wait to try this feature. But I have to. Many thanks again for the demo |
Beta Was this translation helpful? Give feedback.
-
Are you using this in the webui before it's done? How? |
Beta Was this translation helpful? Give feedback.
-
|
You are in GitHub, the code is available if you dare to use it. |
Beta Was this translation helpful? Give feedback.
-
|
I tried this but it had no effect for me even when changing the slider from 1 all the way down to 0. |
Beta Was this translation helpful? Give feedback.
-
|
It only works as far as I know for SD-v1.-5-inpainting.ckpt |
Beta Was this translation helpful? Give feedback.
-
|
I see, that's disappointing, I thought it would affect the img2img pipeline regardless of the model being used. |
Beta Was this translation helpful? Give feedback.
-
You could try merging that model with another one. |
Beta Was this translation helpful? Give feedback.
-
|
Good news: this feature merged today from 86e19fe build! I will try it! |
Beta Was this translation helpful? Give feedback.
-
|
I added inpainting_mask_weight to Quicksettings list, so it's always at the top of your screen. |
Beta Was this translation helpful? Give feedback.
-
|
I have written a quick guide to its use here. It is useful to fix it in the Quicksettings list for your convenience |
Beta Was this translation helpful? Give feedback.
-
|
This is an excellent feature. I would like to see it on the img2img tab. So far, I have been using the xy grid to explore the parameters without having to tab to settings. |
Beta Was this translation helpful? Give feedback.
-
|
when you use inpainting with mask with a value between 0.0 and 1.0 the colors are skewed to white |
Beta Was this translation helpful? Give feedback.
-
Did anyone find a solution to this issue? |
Beta Was this translation helpful? Give feedback.
-
I've been trying this feature (still not released, for SD-v1.-5-inpainting.ckpt), and I can say that is a game changer for img2img and it has a lot of useful applications. The combination with Loopback is amazing.
Many thanks to all the developers of Automatic1111 for your amazing work.
Beta Was this translation helpful? Give feedback.
All reactions