Software Engineering Courses (SECourses)•3y ago

One thing iv noticed with learning rate .000001 is when you generate ur image, we will see preview a

One thing iv noticed with learning rate .000001 is when you generate ur image, we will see preview and it looks like a perfect face from your data set. However, when generating is done they don't actually look like what u see during generating stage. What could be an indication or hope to make this improve. Training in dreambooth still not getting better precise look from ur subject at least 5 images from hundreds of generation clicks

Anyone share your experience?

TrimacOP•3/9/23, 12:51 AM

TrimacOP•3/9/23, 12:53 AM

Iv been testing Yt tutorials and none of them works. It could be a lie configurations and miss leading information to keep us back in their channel's.

FFurkan Gözükara SECourses Click to see attachment

Haashim•3/9/23, 12:55 AM

This looks great. Easy to read

Haashim•3/9/23, 12:56 AM

I may start watching your coding videos lol.

You could probly do a nice class on coding for AI eventually

TTrimac One thing iv noticed with learning rate .000001 is when you generate ur image, w...

dave5555•3/9/23, 1:10 AM

what scheduler you using?

dave5555•3/9/23, 1:11 AM

i have noticed sometime that halfway through it looks promising and then fizzles out haha

dave5555•3/9/23, 1:12 AM

how many images in dataset ? using class images? text encoder?

dave5555•3/9/23, 1:13 AM

i hear that if you are doing a small set where you want to recreate a certain face you should do reg images

dave5555•3/9/23, 1:15 AM

@Furkan Gözükara SECourses how hard should it be to copy and paste code from a colab notebook to run in jupyter lab ipynb notebook

dave5555•3/9/23, 1:16 AM

i was trying to port over the kohya trainer from colab to jupyter and get hung up on the dependency installation , its trying to download google stuff and throws errors, and im a noob but trying to learn

Ddave5555 what scheduler you using?

TrimacOP•3/9/23, 2:08 AM

Almost all scheduler. I think I will move to Lora again

john•3/9/23, 3:23 AM

Hello all

john•3/9/23, 3:23 AM

I am an amateur videographer with creative aspirations

john•3/9/23, 3:24 AM

I have footage I'd like to use to train models

john•3/9/23, 3:24 AM

I have no experience

john•3/9/23, 3:24 AM

I am eager to learn

john•3/9/23, 3:24 AM

Hoping I'm in the right place to make some progress or get pointed in the right direction

P1pSl1p•3/9/23, 3:41 AM

I'm getting really solid results with my LoRA. it works great with close-ups. but starts falling apart on wide shots.

P1pSl1p•3/9/23, 3:42 AM

I tried inpainting the face but the results aren't very good. Is this a result of my dataset being too limited in the number of wide shots that I fed it and therefore SD doesn't know how to draw her face from afar? or is there some settings and techniques I'm missing?

Enlil Wolf•3/9/23, 4:37 AM

Anyone using runpod here knows how to stop a runpod instance that is still running? As it is not allowing me to run a new instance.

EEnlil Wolf Anyone using runpod here knows how to stop a runpod instance that is still runni...

justalurker•3/9/23, 4:58 AM

runpodctl stop pod $RUNPOD_POD_ID

Jjustalurker runpodctl stop pod $RUNPOD_POD_ID

Enlil Wolf•3/9/23, 4:59 AM

Thank you sir

justalurker•3/9/23, 5:00 AM

Or if you want to set it to stop after a certain amount of time - (sleep 2h; runpodctl stop pod $RUNPOD_POD_ID) &

Enlil Wolf•3/9/23, 5:05 AM

Just wanted to stop & restart it because I wanted to upload classification files from PC to the runpod.

So, as a complete newcomer, I may be missing something completely simple but I have been following this tutorial : https://www.youtube.com/watch?v=Bdl-jWR3Ukc and I am at around the 36th minute where we begin training the model. I pretty much completed everything up until this point but I keep getting this error the moment I hit train, the error reads : Please configure some concepts.

And it is weird because I did set everything up in the concepts tab of Dreambooth. Anyone that can help or has some knowledge of what that means.

YouTubeSECourses - Software Engineering Courses

Zero To Hero Stable Diffusion DreamBooth Tutorial By Using Automati...

Our Discord : https://discord.gg/HbqgGaZVmr. The most advanced tutorial of Stable Diffusion Dreambooth Training. 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, Automatic1111 and Google Colab ...

PP1pSl1p I tried inpainting the face but the results aren't very good. Is this a result o...

Ruairi Robinson•3/9/23, 5:17 AM

are you using inpainting resolution correctly? here's a quick tutorial that explains the best method: https://www.youtube.com/watch?v=c8TGgKrr7s8&list=LL&index=30

RRuairi Robinson are you using inpainting resolution correctly? here's a quick tutorial that expl...

P1pSl1p•3/9/23, 5:25 AM

Cheers. I'll take a look after lunch.

PP1pSl1p I'm getting really solid results with my LoRA. it works great with close-ups. b...

TrimacOP•3/9/23, 5:32 AM

This is better style! nice

PP1pSl1p I tried inpainting the face but the results aren't very good. Is this a result o...

TrimacOP•3/9/23, 5:38 AM

I used inpainting to improve my face from best model trained from my face. I used same instance prompts and negative from a specific image inside PNG tab include the settings bellow. It should same 512 or whatever ur size to much the image then inpaint only area

P1pSl1p•3/9/23, 7:01 AM

do LoRAs not work with inpainting?

PP1pSl1p do LoRAs not work with inpainting?

Ruairi Robinson•3/9/23, 7:16 AM

they do

P1pSl1p•3/9/23, 7:17 AM

yeah, I'm starting to see that. slowly.lol

P1pSl1p•3/9/23, 7:18 AM

just a lot of settings tweaking

P1pSl1p•3/9/23, 7:18 AM

getting some weird artifacts tho. like a face inside a head.

P1pSl1p•3/9/23, 7:18 AM

not a face on a face

P1pSl1p•3/9/23, 7:22 AM

cranking up the padding pixels pretty high seems to help tho.

P1pSl1p•3/9/23, 7:22 AM

oh yeah. that did it.

Original message was deleted

Enlil Wolf•3/9/23, 9:21 AM

For dreambooth?

Sibun•3/9/23, 9:53 AM

hello guys
Ive use dreambooth colab to train a model, after convert weights to ckpt, ive use this for Automatic1111 webui (colab) but i cant use this model
When I change the checkpoints into my model, I think this error appear:
Calculating sha256 for /content/drive/.shortcut-targets-by-id/12I1MvuQ-v47Ra9vsoge22RbR2zhXyvU2/StableDiffusion/stable-diffusion-webui/models/Stable-diffusion/model.ckpt: Skipping broken symlink: /content/drive/.shortcut-targets-by-id/12I1MvuQ-v47Ra9vsoge22RbR2zhXyvU2/StableDiffusion/stable-diffusion-webui/models/Stable-diffusion/Checkpoint
aedbb913d928ef2c2af4d51a20b2d5a3a2f2cce1a762801dffc9221d0eb7c1f7
And its frozen at: Loading weights [aedbb913d9] from /content/drive/.shortcut-targets-by-id/12I1MvuQ-v47Ra9vsoge22RbR2zhXyvU2/StableDiffusion/stable-diffusion-webui/models/Stable-diffusion/model.ckpt.
Nothing change, i cant load that checkpoints
How to fix this guys?

gatsby•3/9/23, 10:28 AM

Hey everyone

when using dreambooth and generating class images, is it better to get very high quality images close to the images you’re providing in the dataset images?

Ggatsby Hey everyone 🙂 when using dreambooth and generating class images, is it better ...

P1pSl1p•3/9/23, 10:34 AM

I spent some time curating my images by deleting those that had deformities. Images that didn't match my classification. And generally anything that I didn't want my model to learn, ie. Nudity, etc.

P1pSl1p•3/9/23, 10:34 AM

I had a classification directory of 2.4k images so it took some time. Did it make a difference? I don't know.

PP1pSl1p I spent some time curating my images by deleting those that had deformities. Ima...

gatsby•3/9/23, 10:35 AM

Interesting. So it’s worth having the same quality class images as the dataset training images

gatsby•3/9/23, 10:35 AM

It’s just a matter of if the individual has the time to go generate or input so many class images?

P1pSl1p•3/9/23, 10:35 AM

Can us SD to generate the class images

P1pSl1p•3/9/23, 10:36 AM

Txt2Img

PP1pSl1p Txt2Img

gatsby•3/9/23, 10:36 AM

Yes, have tried that. Thanks for clarifying!

TTrimac One thing iv noticed with learning rate .000001 is when you generate ur image, w...

Furkan Gözükara SECourses•3/9/23, 10:39 AM

this means your prompt is the problem. you need to adjust your prompt. too long prompt usually drives it away from your subject

Ddave5555 @Furkan Gözükara SECourses how hard should it be to copy and paste code from a c...

Furkan Gözükara SECourses•3/9/23, 10:40 AM

should be very easy

Jjohn Hello all

Furkan Gözükara SECourses•3/9/23, 10:41 AM

hello welcome

Furkan Gözükara SECourses•3/9/23, 10:41 AM

first start installing with automatic1111

One thing iv noticed with learning rate .000001 is when you generate ur image, we will see preview a

Similar Threads