[Stable Diffusion] Prompt Sharing and Learning Thread

[theMickey_]

here before and after using reactor.

Seems to me like it's working -- what exactly is your issue? I must say though, the model you're using to generate the "original" image and the image from which you want to use the face have (very) different art styles, so the "after" image is exactly how I'd expect it to be -- the face of the "after" image is not matching the art style of the rest of the image.

If you want to get a better result, maybe try to use the face you want to use for replacement and do an img2img transformations with the same model you're using to generate the "before" image, hopefully to get a better matching art style of just the face. And then use the transformed face in ReActor. Not sure if that's working, but that's how I would try to get a better result.

 

[Sepheyer]

Dudes, would anyone know why Face Restore stopped working in CUI after a long while and if it has a fast fix? Naturally, there are error messages, etc., I am checking to see if someone ran into this very issue, i.e.: oh yea, the custom_node interface got trashed, you need to re-dl this [1, 2, 3]. Else I'll be diving into the error list which clearly is FML.

 

[Synalon]

You mean u want the orginal and

here before and after using reactor.

Upload the image you want the new face to go on, and the picture you are loading into reactor to provide the face.

 

[deadshots2842]

Upload the image you want the new face to go on, and the picture you are loading into reactor to provide the face.

Can you explain with example for auto1111 i didn't understand what you're saying.

 

[Sharinel]

Can you explain with example for auto1111 i didn't understand what you're saying.

What everyone is telling you and what you are ignoring is that in the example images you attached in your previous post, Reactor is working exactly how it should. You just don't like the end result.

Here, I'll even post examples that you demand in every one of your messages

This was the original that I used following your prompt, except for the lora


Now here she is with Reactor adding on the face you wanted.



Your reactor image isn't really compatible with the checkpoint you are using. I then used the following pic as the Reactor face instead



Which gave me this result



It all depends entirely on the image you use for the reactor face. Using a full body picture when all you are cloning is the actual face is not the best method of doing it. Try and find a high resolution image of just the face and shoulders

 

[Synalon]

Can you explain with example for auto1111 i didn't understand what you're saying.

If you upload the 2 images I asked for I will do a screenshot guide for you.

Sharinel is also correct that you will probably need an anime checkpoint to do anime.

Now Upload the picture you want the face on. And upload the picture of the face you want to apply.

 

[theMickey_]

Now Upload the picture you want the face on. And upload the picture of the face you want to apply.

He already posted the original image and the face he wants to swap with. He also already did the actual face swap, so he knows how it's done. He just don't like the result, and we've already discussed why that might be the case (art styles not matching).

 

[Synalon]

He already posted the original image and the face he wants to swap with. He also already did the actual face swap, so he knows how it's done. He just don't like the result, and we've already discussed why that might be the case (art styles not matching).

I see, theres nothing left I can do then.
I added a picture I made using Face ID in controlnet, still doesn't match the face he wanted in the picture though.

 

[Synalon]

Very Basic guide on how to use Reactor

Spoiler: Steps

You don't have permission to view the spoiler content. Log in or register now.

 

[Halmes]

Hey. Whats the best model to use training lora?

 

[Synalon]

Hey. Whats the best model to use training lora?

That would depend on what checkpoint you intend to use it with.

If you want to use it mostly with checkpoints trained on SD1.5 then use SD1.5, if you use mostly XL checkpoints use SDXL.

If you only ever use 1 checkpoint then just train on that.

 

[Halmes]

Im using google colab to train. I have 735 pics. How many repeats and epochs should i do? 10 epochs, 3 batch, 10 repats says too much steps.

 

[devilkkw]

What are you training? Style or subject?
700 is really too many pics.
I'm speaking about sd1.5 . In mostly case i use max 20 image, 18 steps and 4 epoch, for subject, and 36 step and 3 epoch for style.
I'm unable to test sdxl, train sdxl lora on 6gb is impossible, or maybe i don't know the correct way.

 

[felldude]

Im using google colab to train. I have 735 pics. How many repeats and epochs should i do? 10 epochs, 3 batch, 10 repats says too much steps.

I would look at the mean ar error (without repeats) with that many images. If it is in the 100th decimal your probably ok
(I have locally trained high error lora's in the 10ths for ar error, I think I trained 680 images with an AR error of 0.4)

Train on the EMA checkpoint if you can for

You must be registered to see the links

(I can't find the un-pruned EMA checkpoint anymore)
The XL EMA weights I recall being 16GB I can not find the checkpoint anymore.

EMA is more important for refinement then for a LORA.
Unless your training a style, the base EMA model produces a lora with wider range.

You could do a batch size of 4 and probably 4 epochs as your well over 10,000 steps

 

[Sharinel]

PSA regarding Forge, looks like it is no longer being maintained. Pity, as it is my main program to generate images

You must be registered to see the links

 

[Halmes]

I would look at the mean ar error (without repeats) with that many images. If it is in the 100th decimal your probably ok
(I have locally trained high error lora's in the 10ths for ar error, I think I trained 680 images with an AR error of 0.4)

Train on the EMA checkpoint if you can for

You must be registered to see the links

(I can't find the un-pruned EMA checkpoint anymore)
The XL EMA weights I recall being 16GB I can not find the checkpoint anymore.

EMA is more important for refinement then for a LORA.
Unless your training a style, the base EMA model produces a lora with wider range.

You could do a batch size of 4 and probably 4 epochs as your well over 10,000 steps

Im training with this

You must be registered to see the links

 

[Halmes]

I would look at the mean ar error (without repeats) with that many images. If it is in the 100th decimal your probably ok
(I have locally trained high error lora's in the 10ths for ar error, I think I trained 680 images with an AR error of 0.4)

Train on the EMA checkpoint if you can for

You must be registered to see the links

(I can't find the un-pruned EMA checkpoint anymore)
The XL EMA weights I recall being 16GB I can not find the checkpoint anymore.

EMA is more important for refinement then for a LORA.
Unless your training a style, the base EMA model produces a lora with wider range.

You could do a batch size of 4 and probably 4 epochs as your well over 10,000 steps

Shold i change to the sd training model too?

 

[Halmes]

I would look at the mean ar error (without repeats) with that many images. If it is in the 100th decimal your probably ok
(I have locally trained high error lora's in the 10ths for ar error, I think I trained 680 images with an AR error of 0.4)

Train on the EMA checkpoint if you can for

You must be registered to see the links

(I can't find the un-pruned EMA checkpoint anymore)
The XL EMA weights I recall being 16GB I can not find the checkpoint anymore.

EMA is more important for refinement then for a LORA.
Unless your training a style, the base EMA model produces a lora with wider range.

You could do a batch size of 4 and probably 4 epochs as your well over 10,000 steps

Think the problem is the paywall in google colab

 

[felldude]

Think the problem is the paywall in google colab

Based off of the error message your trying to use CPU training and need to

You must be registered to see the links

The drop down your showing is non EMA base checkpoint so in that case I would just use what ever model is closest to the style your going for.

With an image count that high you might want to lower the LR to 1e-6 for the U-NET and 5e-7 for the TE

 

[Halmes]

Based off of the error message your trying to use CPU training and need to

You must be registered to see the links

The drop down your showing is non EMA base checkpoint so in that case I would just use what ever model is closest to the style your going for.

With an image count that high you might want to lower the LR to 1e-6 for the U-NET and 5e-7 for the TE

GPU connection cannot be established due to Colab usage
says this. So i have to pay to use gpu more.
I changed google account but now it says " An NVIDIA GPU may be present on this machine, but a CUDA-enabled jaxlib is not installed. Falling back to cpu. "