Software Engineering Courses (SECourses)•3y ago

would be very usefull to have a list of optimal paremeters for dreambooth traning somewhere, will be

would be very usefull to have a list of optimal paremeters for dreambooth traning somewhere, will be easier to see it compared to searching thorugh the videos

MMarikcraven Hello. New here. Is there a tutorial on what/how to caption dataset for dreamboo...

Furkan Gözükara SECourses•3/11/23, 8:55 PM

welcome

Furkan Gözükara SECourses•3/11/23, 8:55 PM

how to caption them depends on you

Furkan Gözükara SECourses•3/11/23, 8:55 PM

but do you ask how to use captions? filewords?

BBrofessor would be very usefull to have a list of optimal paremeters for dreambooth tranin...

Furkan Gözükara SECourses•3/11/23, 8:55 PM

i suggest not use higher batch size. because it changes how they are getting trained

Furkan Gözükara SECourses•3/11/23, 8:55 PM

for faces 1 and 1 works best

Furkan Gözükara SECourses•3/11/23, 8:56 PM

you watched this one?

Furkan Gözükara SECourses•3/11/23, 8:56 PM

https://youtu.be/sRdtVanSRl4

YouTubeSECourses - Software Engineering Courses

Epic Web UI DreamBooth Update - New Best Settings - 10 Stable Diffu...

RunPod: https://bit.ly/RunPodIO. Discord : https://bit.ly/SECoursesDiscord. Web UI DreamBooth got epic update and we tested all new features to find the best settings (10). If I have been of assistance to you and you would like to show your support for my work, please consider becoming a patron on

https://www.patreon.com/SECourses

Playlist of...

BrofessorOP•3/11/23, 8:56 PM

yeah i noticed, i am using 2x2 and faces are not getting better

Furkan Gözükara SECourses•3/11/23, 8:56 PM

i recently trained a style by using settings in this video

BrofessorOP•3/11/23, 8:56 PM

yes i watched both

BrofessorOP•3/11/23, 8:56 PM

the previous one

Furkan Gözükara SECourses•3/11/23, 8:56 PM

1x1

Furkan Gözükara SECourses•3/11/23, 8:56 PM

it yielded very good results

BrofessorOP•3/11/23, 8:56 PM

just i rented a nice GPU so i thought 2x2 would be better

BrofessorOP•3/11/23, 8:56 PM

yes for your face it was amazing

BBrofessor just i rented a nice GPU so i thought 2x2 would be better

Furkan Gözükara SECourses•3/11/23, 8:57 PM

i think that is necessary when you need high computation power which means fine tuning with thousands of images

BrofessorOP•3/11/23, 8:57 PM

i traned first on anime , and it went ok, but for human face was too hard

BBrofessor i traned first on anime , and it went ok, but for human face was too hard

Furkan Gözükara SECourses•3/11/23, 8:57 PM

really decent quality

BrofessorOP•3/11/23, 8:57 PM

yes, because of special models from civitai

BrofessorOP•3/11/23, 8:57 PM

ok, i will try 1x1 then

Furkan Gözükara SECourses•3/11/23, 8:58 PM

if you want higher quality

Furkan Gözükara SECourses•3/11/23, 8:58 PM

1024x1024 definitely improvement

Furkan Gözükara SECourses•3/11/23, 8:58 PM

if your dataset is such

Furkan Gözükara SECourses•3/11/23, 9:00 PM

i am planning a style teaching video hopefully soon

Furkan Gözükara SECourses•3/11/23, 9:00 PM

like corridor crew

BrofessorOP•3/11/23, 9:04 PM

thats very nice

BrofessorOP•3/11/23, 9:04 PM

There were very little tutorials on dreambooth for runpod

BrofessorOP•3/11/23, 9:05 PM

yours was a lifesaver

BrofessorOP•3/11/23, 9:05 PM

Just would be nice to have a list of parameters, because I am having hard time to find all of them in 1 hour videos

BrofessorOP•3/11/23, 9:05 PM

I watched 2 videos, but just checking which parameters you used would be much faster..

FFurkan Gözükara SECourses i am planning a style teaching video hopefully soon

BrofessorOP•3/11/23, 9:05 PM

Looking forward for it

FFurkan Gözükara SECourses but do you ask how to use captions? filewords?

Marikcraven•3/11/23, 9:07 PM

Yes, and what they are for. I've tried researching and found that it's either to include or exclude information but nothing definitive on which so maybe it's both?

Marikcraven•3/11/23, 9:09 PM

Meaning if I have a data set of images with a yellow car in too many of them I would use captions to tell the ai "yellow car" to make sure it doesn't try to train the yellow car? Hopefully that makes sense.

BrofessorOP•3/11/23, 9:22 PM

if your dataset is not too large, you can just photoshop it out, does not take much time actually

MMarikcraven Yes, and what they are for. I've tried researching and found that it's either to...

Furkan Gözükara SECourses•3/11/23, 9:30 PM

The primary reason is

Marikcraven•3/11/23, 9:30 PM

It's pretty big actually, and I'd also like to be able to teach it a few things for the same model not just a face, or one object.

Furkan Gözükara SECourses•3/11/23, 9:30 PM

Improving knowledge of model

Furkan Gözükara SECourses•3/11/23, 9:31 PM

So when you add a keyword

Furkan Gözükara SECourses•3/11/23, 9:31 PM

And associated with an image with it

Furkan Gözükara SECourses•3/11/23, 9:31 PM

Model gets improved to that new image for that keywords

Furkan Gözükara SECourses•3/11/23, 9:32 PM

In this video I explain how to use filewords

Furkan Gözükara SECourses•3/11/23, 9:32 PM

https://youtu.be/KwxNcGhHuLY

YouTubeSECourses - Software Engineering Courses

DreamBooth Got Buffed - 22 January Update - Much Better Success Tra...

Our Discord : https://discord.gg/HbqgGaZVmr. Newest update of DreamBooth extension of Automatic1111 brought huge quality and success improvement. If I have been of assistance to you and you would like to show your support for my work, please consider becoming a patron on

https://www.patreon.com/SECourses

Playlist of Stable Diffusion Tutorials...

Furkan Gözükara SECourses•3/11/23, 9:32 PM

They are also used to not memorize repetitive things in images but I am yet to test that

Marikcraven•3/11/23, 9:32 PM

Awesome. I'll watch that one right now. Thank you very much for the help.

MMarikcraven Awesome. I'll watch that one right now. Thank you very much for the help.

Furkan Gözükara SECourses•3/11/23, 9:33 PM

I am planning a technical video. Hopefully if I can make it, it will explain a lot

Marikcraven•3/11/23, 9:33 PM

You are awesome. I love your videos keep up the great work!

BrofessorOP•3/11/23, 10:10 PM

actually i tihnk training with higher batch sizes may be ok, but looks like learning rate is still better to be kept at 2e-6

BBrofessor actually i tihnk training with higher batch sizes may be ok, but looks like lear...

Furkan Gözükara SECourses•3/11/23, 10:18 PM

the higher batch size changes how the model learns

Furkan Gözükara SECourses•3/11/23, 10:18 PM

it is called as mini batches vs full batch

Furkan Gözükara SECourses•3/11/23, 10:18 PM

you can look for the effect

would be very usefull to have a list of optimal paremeters for dreambooth traning somewhere, will be

Similar Threads