Software Engineering Courses (SECourses)•3y ago

Hi, Thx much for the dreambooth training tutorial on patreon. I just completed training dreambooth v

Hi, Thx much for the dreambooth training tutorial on patreon. I just completed training dreambooth via Automatic111 using custom images. Post training the text2img inference works perfect on Automatic111 and I am getting good quality image after 6hrs training following your steps. I had one issue though - when I use the same checkpoint on comfyUI as is, I just get random noise. Is the checkpoint generated via dreambooth on Automatic111 missing something?

SgtSnicker•10/20/23, 4:49 AM

I posted a screenshot here yesterday @rafraf 4090, 64 gigs of RAM as well. I was getting near 12 it/s using SDXL and 1024x1024. Unfortunate that you had all those errors. I had to completely reinstall my SD after I messed up generating a new TensorRT for a different SDXL checkpoint. Unfortunately I know nothing about all this stuff and was able to not mess up on my second time. Now I just stick with default rather than 1024 statics lol

DR.Siva•10/20/23, 4:57 AM

Many Lora and Checkpoints need some kind of trigger words to be included. Although not everything. For LORA , we have the ability to add certain words when its added from lora tab , but for checkpoints/Models , i dont see such option. It hugely difficult to remember and add those trigger words. Is there any other way to automate this ?

DR.Siva•10/20/23, 4:58 AM

DR.Siva•10/20/23, 5:19 AM

ok nvm , found . civitai helper extension.

mikemenders•10/20/23, 8:57 AM

Does anyone use Adafactort for concept training? If so, with what parameters? I found two completely different arguments on the net, so I'm getting lost in the woods.

DDR.Siva Many Lora and Checkpoints need some kind of trigger words to be included. Althou...

mikemenders•10/20/23, 9:03 AM

It has been solved in Auto1111. When you open the Lora data sheet, just click on the words you want to load with the lora or type in the Activation text field

mikemenders•10/20/23, 9:05 AM

DR.Siva•10/20/23, 9:05 AM

it works for lora, but more models ?

DR.Siva•10/20/23, 9:06 AM

even though 90% of models dont need them , some specify the trigger word as mentioned in my post

mikemenders•10/20/23, 9:06 AM

There in the Notes field you copy the words

mikemenders•10/20/23, 9:07 AM

DR.Siva•10/20/23, 9:14 AM

I agree. but i have around 10+ SDXL models and play around them, remembering or missing them is not an option right ?

LEGACIES•10/20/23, 10:02 AM

I have trained my model and try to test it but getting this error. Does anyone know how to solve it ?

LEGACIES•10/20/23, 10:02 AM

this is on runpod

mikemenders•10/20/23, 10:10 AM

Your space is full...

LEGACIES•10/20/23, 10:11 AM

i have only trained 1 model

Narva•10/20/23, 10:56 AM

Any tip for make the sdxl imagen generation more fast? take a lot of time generate just one pic, I am using 3090 in runpod btw

LLEGACIES i have only trained 1 model

Ibaraki Douji•10/20/23, 11:33 AM

you still have no disk space left on your pc

Wwip wai Hi, Thx much for the dreambooth training tutorial on patreon. I just completed t...

Furkan Gözükara SECourses•10/20/23, 11:40 AM

this could be due to selected sampler or vae? have you tested different samplers more step count?

DDR.Siva Many Lora and Checkpoints need some kind of trigger words to be included. Althou...

Furkan Gözükara SECourses•10/20/23, 11:40 AM

caption all images good

Furkan Gözükara SECourses•10/20/23, 11:40 AM

you wont need trigger word

Furkan Gözükara SECourses•10/20/23, 11:41 AM

you can use my SOTA captioners

Furkan Gözükara SECourses•10/20/23, 11:41 AM

https://youtu.be/PNA9p94JmtY

YouTubeSECourses

Caption Images or Learn How To Prompt With Clip Vision of SDXL and ...

Download scripts from here : https://www.patreon.com/posts/sota-very-best-90744385

Full length Twitter video here : https://twitter.com/GozukaraFurkan/status/1712665890259447831

SOTA (The Very Best) Image Captioning Models Script For Stable Diffusion And More

1 Click install and use SOTA image captioning models on your computer. Supports 8 bi...

Mmikemenders Does anyone use Adafactort for concept training? If so, with what parameters? I ...

Furkan Gözükara SECourses•10/20/23, 11:42 AM

it works for all. concept training you mean general fine tuning?

LLEGACIES I have trained my model and try to test it but getting this error. Does anyone k...

Furkan Gözükara SECourses•10/20/23, 11:42 AM

yes your runpod is out of space

Furkan Gözükara SECourses•10/20/23, 11:42 AM

turn off pod and increase space or delete some checkpoints

Furkan Gözükara SECourses•10/20/23, 11:42 AM

with

Furkan Gözükara SECourses•10/20/23, 11:42 AM

rm -r filename

NNarva Any tip for make the sdxl imagen generation more fast? take a lot of time gener...

Furkan Gözükara SECourses•10/20/23, 11:42 AM

tensorrt

Furkan Gözükara SECourses•10/20/23, 11:43 AM

hopefully i will make a tutorial for that for runpod too

mikemenders•10/20/23, 11:48 AM

I am happy now. I looked at several of Lora's profiles that used poses, concepts, clothing features. Based on that, I put together the file below. However, I also noticed two tricks:
- they set a 10x repeat regardless of the number of images
- usually for 10 epochs, so the number of images x100 is the measure in most cases.
This time, however, I used WD14 tagging on SD 1.5. I removed the labels that were valid for my concept and a few that could possibly affect the final result of the generated image (e.g. blonde hair, brown eyes), and I left one of the double labels as phone and cellphone.
And indeed, the 10th epoch model became absolutely flexible. And it even works without a keyword! I love it, and it shows how important the tagging and the repetition of images and the number of steps are.

Mykee_-_pose_and_concept.json2.84KB

FFurkan Gözükara SECourses it works for all. concept training you mean general fine tuning?

mikemenders•10/20/23, 11:50 AM

Not just fine tuning, but also a new keyword. I experimented with Onoff model, where you have to learn the model:
- the split screen,
- the person on the left and the person on the right are the same
- the person on the left has clothes on, the person on the right has clothes off, same body shape
- maybe even wear the same pose on both sides of the person
- while remaining flexible on face, dress, background, but background on both sides is fine if the same.

FFurkan Gözükara SECourses caption all images good

DR.Siva•10/20/23, 12:04 PM

Don't understand sir

DDR.Siva Don't understand sir

Furkan Gözükara SECourses•10/20/23, 12:21 PM

ye you need to understand how training happens. long thing to explain. you need to watch all tutorials

Furkan Gözükara SECourses•10/20/23, 12:22 PM

you can also read this paper : https://arxiv.org/abs/2208.12242

arXiv.org

DreamBooth: Fine Tuning Text-to-Image Diffusion Models for Subject-...

Large text-to-image models achieved a remarkable leap in the evolution of AI, enabling high-quality and diverse synthesis of images from a given text prompt. However, these models lack the ability to mimic the appearance of subjects in a given reference set and synthesize novel renditions of them in different contexts. In this work, we present a...

LEGACIES•10/20/23, 12:55 PM

when completing training with dreambooth, why do we merge the trained model with the original model used?

LLEGACIES when completing training with dreambooth, why do we merge the trained model with...

Furkan Gözükara SECourses•10/20/23, 1:28 PM

we dont merge

LEGACIES•10/20/23, 1:35 PM

any reason why some people merge and some do not?

LLEGACIES any reason why some people merge and some do not?

Furkan Gözükara SECourses•10/20/23, 1:43 PM

they could be merging to get other model style

Furkan Gözükara SECourses•10/20/23, 1:43 PM

but no need to merge with base model you trained on

FFurkan Gözükara SECourses you can use my SOTA captioners

Elisoli•10/20/23, 2:12 PM

could you tell us which models did you use to train your lest set for the costumer you said about ealier? when i tried a few models from there it wasnt very accurate. Also, as discussed here before, why use captions when you use regularlazation images?

DDR.Siva even though 90% of models dont need them , some specify the trigger word as ment...

DR.Siva•10/20/23, 2:12 PM

@Dr. Furkan Gözükara Sir are we talking about this