Software Engineering Courses (SECourses)•3y ago

okay, so I can train a person and style into 1 model using the colab 'version' of dreambooth?

FFurkan Gözükara SECourses https://youtu.be/m-UVVY_syP0

WolvenkoningOP•5/9/23, 12:07 PM

thx!!

WWolvenkoning okay, so I can train a person and style into 1 model using the colab 'version' o...

Furkan Gözükara SECourses•5/9/23, 12:25 PM

yes you can

Furkan Gözükara SECourses•5/9/23, 12:25 PM

logic is same

WolvenkoningOP•5/9/23, 12:53 PM

okay cool!

WolvenkoningOP•5/9/23, 12:53 PM

ill try it in a bit! thanks in advance

Spydah•5/9/23, 1:10 PM

If there's a model that already has the style you are interested in, it's easier to just train the person into that model.

WolvenkoningOP•5/9/23, 1:13 PM

yeah good call

WolvenkoningOP•5/9/23, 1:14 PM

for training it. should I write captions for everything? or just the person or the style or even neither one of the 2? what works best?

Spydah•5/9/23, 1:17 PM

If you have a model that already has the style and you are just training a single person into it, you only need to focus on the person. You don't even really need captions. You can just use the instance prompt.

WolvenkoningOP•5/9/23, 1:19 PM

yeah okay. and if I were to train a model from the base 1.5 SD model? is it a good idea to add captions then?

Spydah•5/9/23, 1:19 PM

If you are training multiple things at the same time you'll need to use captions or filewords.

WolvenkoningOP•5/9/23, 1:20 PM

okay thx! ill try that

stavsap•5/9/23, 1:45 PM

managed to run Kohya SS on WSL 2.0

managed to run Lora training but dreambooth i dont have enough VRAM

the web gui is working, just load a model to wsl fs it seems as the default models download not working.

https://github.com/stavsap/sd-docker/blob/main/README.md#kohya-ss-for-wsl-20

GitHub

sd-docker/README.md at main · stavsap/sd-docker

Contribute to stavsap/sd-docker development by creating an account on GitHub.

dominiczy•5/9/23, 1:50 PM

Hi all, I'm trying to use stable diffusion/dreambooth/controlnet for the following purpose: I have one background image (of a room interior) and I want to inpaint a product image (of a furniture product). However I've failed to achieve this so far, always requires text prompts for one or the other rather than just providing an image for both

Spydah•5/9/23, 2:06 PM

Is it a specific piece of furniture?

Spydah•5/9/23, 2:16 PM

Any idea why my lora sample images look like this?

Spydah•5/9/23, 2:20 PM

The model is not usable.

Spydah•5/9/23, 2:22 PM

Thats 20 inputs after 5800 steps

Spydah•5/9/23, 2:24 PM

Loss is not going down either.

Spydah•5/9/23, 2:47 PM

Ok, it's something with my setup, not just lora.

ivaxsirc -4090•5/9/23, 2:49 PM

One question, Dr. Furkan, the big problem we have today is the large volume of photos we have on our hard drives, mainly personal images that cannot be deleted or lost. If we created a specific caption text for each image, could we create a model for stable diffusion with all our trained images, and putting the specific prompt would generate the same image that we have original? I don't know if you understand the question, thank you.

kingPanda8•5/9/23, 6:09 PM

hi i have a question. im following the kohya tutorial and got stuck only 3 mins in. anyway idk where he cloned his files at but i don't have an f drive and i dont have anything that he has in the folder. pls help, i cant move on from this step.

Furkan Gözükara SECourses•5/9/23, 7:12 PM

sorry guys i have been away

Furkan Gözükara SECourses•5/9/23, 7:12 PM

now will check messages

Sstavsap managed to run Kohya SS on WSL 2.0 managed to run Lora training but dreambooth ...

Furkan Gözükara SECourses•5/9/23, 7:13 PM

why do you use wsl instead of native python on windows?

Ddominiczy Hi all, I'm trying to use stable diffusion/dreambooth/controlnet for the followi...

Furkan Gözükara SECourses•5/9/23, 7:13 PM

ye you have to train image into model

Furkan Gözükara SECourses•5/9/23, 7:14 PM

controlnet can also help you with shapes

Furkan Gözükara SECourses•5/9/23, 7:14 PM

but not the original image itself

SSpydah Loss is not going down either.

Furkan Gözükara SECourses•5/9/23, 7:14 PM

loss is not important with dreambooth

Furkan Gözükara SECourses•5/9/23, 7:14 PM

i mean stable diffusion

Furkan Gözükara SECourses•5/9/23, 7:14 PM

not like other mls

Iivaxsirc -4090 One question, Dr. Furkan, the big problem we have today is the large volume of p...

Furkan Gözükara SECourses•5/9/23, 7:14 PM

ah for backing up imageS?

Furkan Gözükara SECourses•5/9/23, 7:14 PM

nope not a good idea

KkingPanda8 hi i have a question. im following the kohya tutorial and got stuck only 3 mins ...

Furkan Gözükara SECourses•5/9/23, 7:15 PM

you need to use your own hard drive

Furkan Gözükara SECourses•5/9/23, 7:15 PM

but looks like you really need help. if you become our patreon supporter i can help you via google remote

FFurkan Gözükara SECourses loss is not important with dreambooth

Spydah•5/9/23, 7:15 PM

ok, maybe more training time then....

FFurkan Gözükara SECourses ah for backing up imageS?

ivaxsirc -4090•5/9/23, 8:06 PM

Right as a backup of your photos, if you want a photo write the prompt and generate so you have a file of 2 GB that could generate thousands of original photos.
It's a shame that can't be done.
Thanks

Spydah•5/9/23, 8:08 PM

FWIW I did get Kohya GUI working on runpod using those instructions I posted yesterday. Just to find out that I prefer to do it with Dreambooth Extension in SD UI. But for those who care, it is possible. https://discordapp.com/channels/772774097734074388/1042802773365891173/1105245748414660749

Spydah•5/9/23, 8:10 PM

@ivaxsirc -4090 You could kind of do it. The generated photos would never look exactly like the originals. And you could make arbitrary photos of situations that never happened. But, it would completely remove the sentimentality of capturing the real moment the photo was taken. Besides, digital storage is relatively cheap.

Elsa•5/9/23, 8:15 PM

Hi everybody, I am trying the "Epic Web UI dreambooth Update" turorial but I am having an error at the 6minute35 of the tutorial. This is the ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
tb-nightly 2.13.0a20230414 requires google-auth-oauthlib<1.1,>=0.5, but you have google-auth-oauthlib 0.4.6 which is incompatible.
open-clip-torch 2.7.0 requires protobuf==3.20.0, but you have protobuf 4.23.0 which is incompatible.

I had no problems when I tried the tutorial " How to install Dreambooth on runpod" and could get the link to run dreambooth via runpod.
But can't with this new tutorial.. Thank you very much in advance

FFurkan Gözükara SECourses why do you use wsl instead of native python on windows?

stavsap•5/9/23, 8:33 PM

i noticed better performance in linux drivers. speed and VRAM utilization

Spydah•5/9/23, 8:43 PM

@Elsa The fastest way to get going on runpod is with this. https://github.com/rogerb831/runpod-dreambooth-easy - It sets your environment exactly like the on in this video https://youtu.be/zA4LksIVas8

GitHub

GitHub - rogerb831/runpod-dreambooth-easy

Contribute to rogerb831/runpod-dreambooth-easy development by creating an account on GitHub.

YouTubeSECourses

RunPod Fix For DreamBooth & xFormers - How To Use Automatic1111 Web...

RunPod: https://bit.ly/RunPodIO. Discord : https://bit.ly/SECoursesDiscord. How to install latest version of Automatic1111 Web UI and #DreamBooth on #RunPod. 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 #StableDiffusi...

RunPod Fix For DreamBooth & xFormers - How To Use Automatic1111 Web...

FFurkan Gözükara SECourses thanks i will check

m5k10•5/9/23, 8:53 PM

Why do you use specific commit from sd_dreambooth_extension ? \

Mm5k10 Why do you use specific commit from sd_dreambooth_extension ? \

Furkan Gözükara SECourses•5/9/23, 8:57 PM

in which video?

Furkan Gözükara SECourses•5/9/23, 8:57 PM

sometimes extension were getting broken

Furkan Gözükara SECourses•5/9/23, 8:57 PM

currently you can use last version

m5k10•5/9/23, 8:57 PM

in this file, you checkout to specific commit https://github.com/rogerb831/runpod-dreambooth-easy/blob/main/runpod-dreambooth.sh

GitHub

runpod-dreambooth-easy/runpod-dreambooth.sh at main · rogerb831/run...

Contribute to rogerb831/runpod-dreambooth-easy development by creating an account on GitHub.

Spydah•5/9/23, 8:59 PM

The GUI layout of both SD UI and Dreambooth Ex. change a lot, and theres a lot of dependency changes also. Using specific commits ensures it just works and looks like the instructional videos

SSpydah The GUI layout of both SD UI and Dreambooth Ex. change a lot, and theres a lot o...

Furkan Gözükara SECourses•5/9/23, 9:05 PM

ye that is being logical

okay, so I can train a person and style into 1 model using the colab 'version' of dreambooth?

Similar Threads

Similar Threads

Similar Threads