Software Engineering Courses (SECourses)•3y ago

These are interesting things. In the last few Lora's I've coached women, I never adjusted the Lora's

These are interesting things. In the last few Lora's I've coached women, I never adjusted the Lora's weight, it was always 1 and never modify CFG (always on 7). I had to use the Hires Fix to make the face even better, but many times it brought the person up nicely without it. But I'm tired of repeating the parameters, if you're interested, check out my comments here.

mikemendersOP•11/28/23, 4:25 PM

Yes, it's Lora, not Dreambooth, and I made it from plain Instagram pictures.

mikemendersOP•11/28/23, 4:30 PM

So a good Lora is not weak

mikemendersOP•11/28/23, 4:38 PM

But this is the end of my proof, I feel I have found the right Lora parameters, so good luck to all Lora tryers!

DR.Siva•11/28/23, 4:59 PM

pretty good, i havent activated teh hi-res for now

DR.Siva•11/28/23, 5:01 PM

this might sound very weird , but i feel automatic1111 starts to give better results after some image creation.. like it warms up after some images.. this is totally absurd thought.. just my feeling..

DR.Siva•11/28/23, 5:06 PM

Also if you are getting abnormal images , switch to 1024x1024 , it will start to give better results as its the base res. Also restarting auto1111 is also a not bad idea.

Mmikemenders These are interesting things. In the last few Lora's I've coached women, I never...

DR.Siva•11/28/23, 5:08 PM

what are your hires settinggs ?

aimademerich•11/28/23, 5:24 PM

Hello I ran into this issue during the kaggle install, ---------------------------------------------------------------------------
ModuleNotFoundError Traceback (most recent call last)
Cell In[2], line 8
5 import threading
7 from flask import Flask
----> 8 from pyngrok import ngrok, conf
10 conf.get_default().auth_token = "---"
12 os.environ["FLASK_ENV"] = "development"

ModuleNotFoundError: No module named 'pyngrok'

[ ]:

!mkdir -p /kaggle/temp/models any suggestions?

aimademerich•11/28/23, 5:24 PM

i placed the token key in and this was the outcome

DDR.Siva what are your hires settinggs ?

mikemendersOP•11/28/23, 5:25 PM

My settings for Hires Fix: 15 steps with 0.4 noise and 4x_NMKD-Superscale-SP_178000_G, but the images above were taken without a hires fix or adetailer, I simply generated them.

Mmikemenders But this is the end of my proof, I feel I have found the right Lora parameters, ...

Benjamin•11/28/23, 7:09 PM

what are your next steps now?

Original message was deleted

AIGambino•11/28/23, 7:09 PM

what does your dataset look like for the face? I cannot get such a clear output with my dataset

BBenjamin what are your next steps now? 🙂

mikemendersOP•11/28/23, 7:20 PM

I have already trained in clothes and style. Prodigy has worked really well for me and I now know what to look out for on Tensorboard to get a good result. I'd like to make a fairly complex conceptual Lora. It's been on hold for a few weeks, but based on the experience of the last training sessions I might start again, but there's no urgency. It will also be a NSFW Lora, partly done by others, but not on a photo basis, but with animation. We'll see how it goes. It works as a separate Lora now, the challenge here will be the balance of kneading 4-5 concepts into one.

AAIGambino what does your dataset look like for the face? I cannot get such a clear output ...

mikemendersOP•11/28/23, 7:22 PM

To prepare a good image file, follow these steps:

Cut out the persons using Furkan's Yolo script.
Use CompreFace to select 25 images with a score of 0.99-1
Use Furkan's scaling/cropping script to create 512x512, 768x768 and 1024x1024 versions.
Train the model with Furkan's reg images, without caption.
With these you should get an absolutely good Lora.

mikemendersOP•11/28/23, 7:30 PM

And the lesson should not be lost:

you don't need many repetitions, 4-5 at most
not many pictures, 20-25, but I have trained style with 10.
the model is intelligent. From the reg pictures it knows it's a woman and from there it knows that women wear clothes or other things. So the model knows that the woman is what I'm training, so it takes her out of their environment. And because it knows the subject is a woman, it learns quickly.
When labelling, you have to train on a model that understands labels (otherwise the model will learn what you don't want them to learn)

mikemendersOP•11/28/23, 7:31 PM

Really unnecessary 10 or more repetitions. You're wasting your time if you're training Lora with lots of reps.

mikemendersOP•11/28/23, 7:34 PM

The reason I showed you the sample images is that it is possible to generate detailed images at 512x512 on SD 1.5 basis, which is faithful to the images you have loaded, without any help from Hires or Adetailer. Of course, the helpers are nice, but even the basic is good.

mikemendersOP•11/28/23, 7:35 PM

And with an SD 1.5 Lora, you can train with almost any smaller video card at home, with up to 6-8 GB of VRAM.

mikemendersOP•11/28/23, 7:37 PM

Yes, CompreFace is also important if you are training with images of strangers, because if you have images with a score of less than 0.99, your model will be bad. I very rarely let you train images with a score of 0.98 (if nothing else), because then the final results will also be scored 0.9 or below on close-up images.

mikemendersOP•11/28/23, 7:39 PM

And the script I posted last time is really good. I also trained with 64/64 weights on the Lora generating the images above. And I used it for styling as well. Just by setting d_coef I even trained SD 1.5 model with it to 768 px.

mikemendersOP•11/28/23, 7:41 PM

I look at the Tensorboard and I can see on the fly, even without sample images, that the d_coef value needs to be set stronger or weaker, and how strong the resulting Lora will be.

mikemendersOP•11/28/23, 7:42 PM

And Prodigy rewards good source images in return for results like this.

mikemendersOP•11/28/23, 7:45 PM

One more tip for CompreFace: choose a source image with both eyes visible and at least an upper body or closer face, and nothing covering the face (sunglasses, hands, whatever). And most important: whatever you want to generate. Since a person can have multiple faces (depending on age, environment), choose the image that is most typical of the person you would generate. Compare the others to that.

mikemendersOP•11/28/23, 7:48 PM

That's it, these are the tricks for a good result.

mikemendersOP•11/28/23, 7:58 PM

Fresh news for personal use only:
https://huggingface.co/stabilityai/sdxl-turbo

stabilityai/sdxl-turbo · Hugging Face

mikemendersOP•11/28/23, 7:59 PM

Test here:
https://clipdrop.co/stable-diffusion-turbo

Clipdrop - SDXL Turbo

mikemendersOP•11/28/23, 8:04 PM

"The generated images are of a fixed resolution (512x512 pix), and the model does not achieve perfect photorealism."

mikemendersOP•11/28/23, 8:05 PM

And on GitHub:
https://github.com/Stability-AI/generative-models

GitHub

GitHub - Stability-AI/generative-models: Generative Models by Stabi...

Generative Models by Stability AI. Contribute to Stability-AI/generative-models development by creating an account on GitHub.

schananas•11/28/23, 9:14 PM

Hi, does someone know if I can train dreambooth inpainting model in automatic1111 or kohya?

r1•11/28/23, 9:25 PM

your 24gb config disables xformers... isnt that super slow?

Mmikemenders To prepare a good image file, follow these steps: - Cut out the persons using Fu...

AIGambino•11/28/23, 9:26 PM

But will this know to center my person in the photo? It wont crop their head off?

AAIGambino But will this know to center my person in the photo? It wont crop their head off...

mikemendersOP•11/28/23, 9:32 PM

Furkan's script for Yolo does not cut the heads off, but cuts the people out of the picture first, so they are in focus. And the cropping script cuts out the images absolutely fine, so you don't have to worry about that, they help a lot in creating the perfect training set.

AIGambino•11/28/23, 9:32 PM

Does he discuss his scripts in a video by chance?

AAIGambino Does he discuss his scripts in a video by chance?

mikemendersOP•11/28/23, 9:38 PM

Here is his cropper script:
https://www.patreon.com/posts/sota-subject-and-88391247

mikemendersOP•11/28/23, 9:39 PM

This include yolo script too
And here is video:
https://www.youtube.com/watch?v=Fbuyu35TkE4

YouTubeSECourses

SOTA Image PreProcessing Scripts For Stable Diffusion Training - Au...

One of the most important aspect of Stable Diffusion training is the preparation of training images. In this tutorial video I will show you how to fully automatically preprocess training images with perfect zoom, crop and resize. These scripts will hugely improve your training success and accuracy.

Scripts Download Link

https://www.patreon...

Sschananas Hi, does someone know if I can train dreambooth inpainting model in automatic111...

Maelstromboni•11/28/23, 9:39 PM

If I understand you OK, it sounds like you want to switch the model used during Inpainting from the first model that generated the base image to a different, specialized model generated from Dreambooth.

If that's the case, you should be able to switch out models during Inpainting to use the other model OK.

Stefanoid9343•11/28/23, 10:23 PM

@Dr. Furkan Gözükara for the pix art the runpod script works now?

Gerdes of Earth•11/28/23, 11:35 PM

Anyone know how to get the most recent version of dreambooth extention working on runpod with A1111?

Gerdes of Earth•11/29/23, 12:18 AM

nevermind I figured it out

Mmikemenders Fresh news for personal use only: https://huggingface.co/stabilityai/sdxl-turbo

Furkan Gözükara SECourses•11/29/23, 1:12 AM

it looks lower quality

Furkan Gözükara SECourses•11/29/23, 1:16 AM

haha

Furkan Gözükara SECourses•11/29/23, 1:18 AM

https://www.linkedin.com/posts/furkangozukara_sdxl-turbo-all-you-need-to-know-about-quality-activity-7135436734695153664-cLB2?utm_source=share&utm_medium=member_desktop

Furkan Gözükara on LinkedIn: SDXL Turbo all you need to know about ...

SDXL Turbo all you need to know about quality

By the way SDXL 20 steps vs 50 steps has 0 difference in terms of quality in like 99% of cases

Sschananas Hi, does someone know if I can train dreambooth inpainting model in automatic111...

Furkan Gözükara SECourses•11/29/23, 1:18 AM

sadly i dont know. i havent seen anywhere in documentations

Rr1 your 24gb config disables xformers... isnt that super slow?

Furkan Gözükara SECourses•11/29/23, 1:18 AM

yes slower but better quality

SStefanoid9343 @Dr. Furkan Gözükara for the pix art the runpod script works now?

Furkan Gözükara SECourses•11/29/23, 1:19 AM

yes working

Furkan Gözükara SECourses•11/29/23, 1:19 AM

i recorded video editing

Furkan Gözükara SECourses•11/29/23, 1:19 AM

improved gui as well

AIGambino•11/29/23, 1:26 AM

Does anyone know why I'm getting tons of weird malformed limbs?

AIGambino•11/29/23, 1:26 AM

I'm using these negative prompts
bad anatomy, bad hands, three hands, three legs, bad arms, missing legs, missing arms, poorly drawn face, bad face, fused face, cloned face, worst face, three crus, extra crus, fused crus, worst feet, three feet, fused feet, fused thigh, three thigh, fused thigh, extra thigh, worst thigh, missing fingers, extra fingers, ugly fingers, long fingers, horn, extra eyes, huge eyes, 2girl, amputation, disconnected limbs, cartoon, cg, 3d, unreal, animate

AAIGambino Does anyone know why I'm getting tons of weird malformed limbs?

Furkan Gözükara SECourses•11/29/23, 1:27 AM

sdxl 1024 x1024?

These are interesting things. In the last few Lora's I've coached women, I never adjusted the Lora's

Similar Threads