SDXL 1.0 Dreambooth LoRA Fine-Tuning!

Рет қаралды 11,197

9 ай бұрын

Resources:
AutoTrain Advanced - github.com/huggingface/autotr...
Dreambooth Training Colab - colab.research.google.com/git...
Also, Abhishek Thakur has a cool YT Channel - @abhishekkrthakur
❤️ If you want to support the channel ❤️
Support here:
Patreon - / 1littlecoder
Ko-Fi - ko-fi.com/1littlecoder

Пікірлер: 58

@behrampatel4872 2 ай бұрын

Thanks for the tutorial. Is there a way to get the collab notebook to use your own local GPU so the training can run faster ? Thanks

@learnweb3603 9 ай бұрын

Bro, do run the refiner as well, can't we load only refiner or is it mandatory to load both base and refiner, because when I am just generating refiner and commenting out the image line while prompt giving me an error Image not defined. Could you please explain this Thank you

5 ай бұрын

these safetensors lora is not recognizing by Automatic111 nor ComfyUI, cant see them in lora tab. What to do? should we convert this dreambooth to normal sdxl or so?

@360_SA 9 ай бұрын

great video can I use it locally if I have powerful GPU. Can I train it for example building for an architect

@flamescales7422 9 ай бұрын

Thanks man, i was always looking for a Google colab notebook for training lora And how do I train in "style"?

@learnweb3603 9 ай бұрын

Amazing job bro , testing it out

@bfvto 3 ай бұрын

Thanks for the video. Do you know if you can use this Lora model with Automactic1111 ? I can’t seem to get it work

@pushakargaikwad9402 9 ай бұрын

Is the trained model a LoRA file that we put inside lora folder in Automatic1111 or is it a standalone complete model checkpoint?

@1littlecoder 9 ай бұрын

It's a Lora file. You need to put inside lora folder.

@pushakargaikwad9402 9 ай бұрын

@@1littlecoder Thanks

@bfvto 3 ай бұрын

did you get it working with A1111 ? I put it in the lora folder, but doesn't work@@1littlecoder

@luciolrv 8 ай бұрын

Great video! Have you seen any project that includes two sks classes together, like sks person and sks dog? I have seen in HuggingFace some datasets that include two different folders of datasets, but it is not clear what they are used for.

@1littlecoder 8 ай бұрын

They use LoRA merge to to merge two concepts. Look for LoRA roulette on Hugging Face spaces

@user-bf8rs6zh7l 4 ай бұрын

how you have manage the issues like this installing diiferent versions of typing-extensions

@CreatingUtopia 3 ай бұрын

yes I am also getting this error while requirement installation

@estrangeiroemtodaparte 9 ай бұрын

Hey, mate. What resolution did you use for the input images? Thanks!

@1littlecoder 9 ай бұрын

I tried to use any sort of square images. But 1024 should be ideal.

@estrangeiroemtodaparte 9 ай бұрын

@@1littlecoder Awesome, thanks!!

@iRushab 6 ай бұрын

Bor after the pytorch_lora_weights.safetensors is created and I hit the Inference play button, nothing happens. It just says 0 sec and a green tick pops up but I cannot find the newly generated images.

@soumyadeephalder6505 4 ай бұрын

remove those "#" at the starting of all the lines

@Tripp111 9 ай бұрын

Thank you! ❤️🍕

@1littlecoder 9 ай бұрын

You're welcome

@rashedulkabir6227 9 ай бұрын

Can a person generate unlimited images using this method?

@sekhardhana3453 9 ай бұрын

Awesome bro

@1littlecoder 9 ай бұрын

Thanks 🤗

@vishalpanjeta2176 9 ай бұрын

How many images do you think should be enough to generate decent photorealistic images?

@xiaojinyusaudiobookswebnov4951 9 ай бұрын

I think about 10-20 depends on how different your face is. (It will better recognize western europoean faces more than say asian faces)

@user-rq2th4nj3h 9 ай бұрын

Can you give me your training data?@@xiaojinyusaudiobookswebnov4951

@altahookah 6 ай бұрын

Hi! it shows this error: 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. lida 0.0.10 requires kaleido, which is not installed. tensorflow-metadata 1.14.0 requires protobuf=3.20.3, but you have protobuf 4.23.4 which is incompatible. tensorflow-probability 0.22.0 requires typing-extensions

@shlokgupta4939 6 ай бұрын

is it fixed yet? I am facing the same issue.

@theflowwizard 6 ай бұрын

I'm also running against these dependency issues. If I continue the training fails with: File "/usr/local/lib/python3.10/dist-packages/autotrain/utils.py", line 280, in wrapper

@CreatingUtopia 3 ай бұрын

any solution?

@tanmaysharma6610 9 ай бұрын

I have trained the model, but how do I run it now to create new images with my prompt?

@1littlecoder 9 ай бұрын

The code is there in the colab. If you use the LoRA adapter then you can use the invoke token like sks ans start prompting

@CreatingUtopia 3 ай бұрын

Hello Bro can you help me run this code? I am getting some unexpected errror like invalid argument gradient checkpoint etc

@stepphen_ffc5606 9 ай бұрын

Can i train clothes?

@walidflux 9 ай бұрын

the whole oint of sdlx is using the refiner after the generation and if you try to refin your images you will lose the the person you trained because he is not on the refiner model

@walidflux 9 ай бұрын

nevertheless thanks for sharing

@sdgaming5132 8 ай бұрын

Bro I try for 6hr but I fail can You Pls make One image for me for Uploading one linkedin black suit image . I will send You my Image on online platform can You pls do it for me

@sanijurrhman4663 9 ай бұрын

What pc requirements for this?

@un_know_n 9 ай бұрын

just something with good gpu with 6gb or more vram and it might take 20 hour or even 30 kin depending upon how powerfull your GPU is

@un_know_n 9 ай бұрын

but you'll h be using in collab then you just need good internet and nothing else as eveything is done on cloud so if youre planing to use collab then i highly suggest you to do so

@wpahp 9 ай бұрын

when I launch inference nothing is happening

@1littlecoder 9 ай бұрын

Did you launch with both the base and refiner models or only base model ? Did you see if there's any picture with created in your current folder

@wpahp 9 ай бұрын

@@1littlecoder i just followed your guide didn’t choose anything I think just trained the lora. Got the safetensor but nothing happens with the inference, although in your guide it looks different, values can be changed etc

@1littlecoder 9 ай бұрын

Did your inference code finish running?

@wpahp 9 ай бұрын

@@1littlecoder yeah it takes like a second

@1littlecoder 9 ай бұрын

In the left hand side (files and folder) is there something called generated image .

@chaowrestaurant1445 9 ай бұрын

Pls make a video on implementing MetaGPT

@huzaifaali5245 9 ай бұрын

error bro

@1littlecoder 9 ай бұрын

What's the error ?

@abhinav-wk3rp 9 ай бұрын

Error

@abhinav-wk3rp 9 ай бұрын

dist-packages/autotrain/cli/run_dreambooth.py", line 451, in __init__ raise ValueError("❌ Please specify a valid image directory") ValueError: ❌ Please specify a valid image directory

@HimanshuKumar-xh2ws 9 ай бұрын

same

@homomanushya3737 8 ай бұрын

for some reason it is not loading the final image it is showing this: OutOfMemoryError Traceback (most recent call last) in () 27 seed = 42 28 generator = torch.Generator("cuda").manual_seed(seed) ---> 29 image = pipe(prompt=prompt, generator=generator).images[0] 30 image = refiner(prompt=prompt, generator=generator, image=image).images[0] 31 image.save(f"generated_image.png")

@1littlecoder 8 ай бұрын

I think you need to clear the memory and then redo the inference after downloading the model