How to train an AI (Stable Diffusion) with your own images (FOR FREE & NO APP)

IMPORTANT: when using the colab, please click 'File' and 'Save a copy in Drive' first to use your own copy.
UPDATE:
A new step by step guide on github:
github.com/Excalibro1/fast-st...
Link to colab:
colab.research.google.com/git...
Inspiration for prompts:
lexica.art/

[@Blues-House]

Great video thanks. Very important to run this from a copy of the collab in your own drive as you say, or you'll lose the trained file. Thumbs up.

[@1mareck1]

you are a legend! been trying to run dreambooth an entire day, then I found this video!! thank you

[@muhammadkalim6946]

100% your are the man and youtube legend that help like a greek fabulous ! Thumbs up...

[@warleroux]

Awesome. I was wondering how to train this thing. Appreciation.

[@younesdom]

Glad to help

[@M3STERL3G3ND]

thank's for this bro. hugs from Brazil.

[@schlib]

It really worked! Thanks 🙏

[@caiocbrunofb7670]

Works! Thanks from Brazil

[@furkankesgin1913]

Thank you for tutorial. I have a problem with "Test the trained model" session. "It seems that you did not perform training during this session or you chose to use a custom path, provide the full path to the model (including the name of the model)" i see this when i tried to start session. Do you know anything about this? Thanks.

[@KatsPurr]

Thanks so much for this tutorial! It worked!

[@crisicleop4783]

hello, i cant see the folder in google drive, where it is supposed to be saved: "DONE, the CKPT model is in your Gdrive in the sessions folder" ...anyone can tell me why i dont find this folder?

[@SmadarBarYochai]

Thank you for this! very helpful and straightforward! I don't understand where in this case you teach the model that those images are of "mansa musa" and not any other person - do the file names matter? do you set it in a different place? thanks again :)

[@younesdom]

Yes exactly its only because of the filenames. They are like the labels for the images you use for training, so generic filenames wouldnt work. You are very welcome :D

[@javiflaja4063]

the notebook doesnt have a field to enter the hugging face access token...

[@muhammadsajid8608]

I'm not getting the Hugging face access token field to paste mine.

[@jords8299]

I think the last step ''testing model'' is not working anymore...

[@shalinidm8089]

im getting error "HTTP Error 401: Unauthorized" in the model download cell. anyone help, please

[@muzaffersozdemir5134]

same here

[@jk5kwastaken]

great video. a couple qeustions: one it keeps saying that it does not know what pyngrok is. and how would I run the ckpt files from google drive. also now in my google drive there are a bunch of random files and pictures

[@Codename-B]

as per today, this video does not follow colab updated features, It will work until the test feature.

[@BilSchimchelevitz]

what about if I enter all my drawings (like more than 500 of them totally different) will it learn my drawing style at the end???

[@krishnamishra2465]

not really

[@personlaura5123]

Hello, thank you very much for this easy-to-use tutorial. I do have one question. I can generate a first image after training my model. However, I can only generate one image, then my session is disconnected. When I try to reconnect, I get this error: "It seems that you did not perform training during this session or you chose to use a custom path, provide the full path to the model (including the name of the model):". However, the training was indeed performed. If I add the name of my model (laura), it doesn't work and I'm told: "The model doesn't exist on you Gdrive, use the file explorer to get the path:". When I try to write the path (../FastDream-booth/Sessions/laura), it doesn't work. Any idea how to fix this error?

[@younesdom]

Sorry never encountered this error, anyone?

[@KelvinMandriaza]

Thhx buddy, helped a lot

[@younesdom]

Glad to hear that!

[@javiflaja4063]

many things are not covered in this video. I have been trying to run this for 3 hrs and nothing yet, I get stcuk at different stages.

[@younesdom]

Sorry to hear that, I did this just for fun on the side ... worked for me :)

[@addisonknox5052]

I think KZbin would love to see a tutorial that has more to it. Even if it's outdated in a year, something 1 year old but will made will still guide those who are lost. This just feels like footnotes.

[@addisonknox5052]

That all is coming from someone who has it running natively on their desktop. And that definitely seemed like a lot more steps than this, but we had a well made guide for that. Having more accessibility would be great, but I got stuck following your lead and I think the probability of you knowing that a huge percent of your audience also got stuck is high. It might be worth reviewing this video and revisiting the subject.

[@addisonknox5052]

The colab has changed over the year and your video is out of date. Even still, you glanced over downloading the model in the set up process and gave a vague mention of hugging face being a resource but no more that. Break it down man

[@NejW.]

Would this work with tensor art?

[@daniels571]

Thanks a lot bro. How do you let the AI know how your model is called? Is it via the jpg name file (e.g. "name.jpg")? or when you give your session a name? I'd really appreciate help :)

[@younesdom]

The model file should be the session name. Your welcome :D

[@daniels571]

@@younesdom thanks :) is it possible to load multiple models in a single session? I tried to move one model of a session in another one (gdrive) but that didnt work

[@LeoAr37]

ty so much!

[@thelbas5128]

i have a question. if i train the model with a name (eg: dave) and images of dave and then after a few days if i train the same model with some different images of dave using the same name, will it overwrite the previous files or will it merge?

[@younesdom]

it will merge and probably create some weird mix between the two daves

[@DickyG41]

Thanks for the video. I seem to have some trouble getting it to work. I get to the point where I upload images, and I keep getting the same error: "RangeError: Maximum call stack size exceeded". The images are all 512x512 jpgs. Have tried numerous times and done some reformatting, but no luck. Any advice is appreciated, thanks!

[@KendoruSutautu]

same

[@KendoruSutautu]

some people say to try chrome (if you're using safari on a mac)

[@younesdom]

Also found out on stackoverflow that this might be because of Safari, if so pls try chrome

[@realtyrone]

So I did this and it worked. I have one question though, is this only for faces? Or can I train other stuff using this?

[@younesdom]

not sure actually cause it seems like dreambooth was made for faces

[@gogidolim]

I installed Easy Diffusion. How do I train it with images I own?

[@NeVeREire]

If I want to use the model I've trained at a later date, how to I load it back up?

[@younesdom]

In the Step Create/Load a Session there's an input 'Session_link_optional' to enter the name of your previous session (which is stored on your google drive)

[@yashwanth4395]

I have trained the model, but while testing , the code is not working

[@thedaviesguitars]

I get a 401 error when I try to run the model download, I add my API key as in my settings but it's not working. Anyone experienced this?

[@shalinidm8089]

did you find solution for this?

[@ExVatic]

I'm getting this error: [Errno 2] No such file or directory: '/content/stable-diffusion-v1-5/text_encoder/pytorch_model.bin at the training cell, any idea what's going on there?

[@ExVatic]

I guess I'm supposed to have stable diffusion installed already?

[@younesdom]

Looks like the previous step to download the model from hugging face didn't succeed. Maybe try a new token from hugging face. You dont need to install anything on your laptop, it all runs in google colab.

[@ExVatic]

@@younesdom thanks will try :)

[@barathelango267]

in the video you used tokens but in the code there is no token should I skip it?

[@achilleschin]

yup. just skip that part

[@JohnGilbertmoore]

Mansa Musa baby!

[@boucheromain-alexis9285]

Hi great video thanks! Do you know how to use this via api? I'd like to plug it to an automation sofware... Any advice would be greatly appreciated :)

[@younesdom]

you can train your model and push it on huggingface, then you can use it via the huggingface API (probably the easiest way)

[@ewilner]

It did not give me an option to upload images. can you help?

[@younesdom]

When you run the "Instance Images" step, there should appear be a little upload button below

[@babejomb8554]

Guys i need an answers , i got 4vram and i installed sd on my computer but i can't training models locally , can i training models on google colab and use the results on my local sd? Will training in Google Colab affect on my vram?

[@mansamusa559]

Training on google colab will not affect your vram and yes u can use the result on ur machine, result is stored in ur google drive

[@babejomb8554]

@@mansamusa559 love u ❤️

[@IJOGijog]

does anyone else have the same problem i keep trying to run the model cell after i paste the hugging face token and it says error, it is only just now doing this even tho i have used the doc before and have already accepted the terms

[@younesdom]

Idk why but something changed with the hugging face API. I ran it yesterday without a token. Try to copy the nb from here colab.research.google.com/github/TheLastBen/fast-stable-diffusion/blob/main/fast-DreamBooth.ipynb again and it should run without a token

[@IJOGijog]

@@younesdom thank you bro it worked!

[@cgc2300]

hello would this allow me to place the images of my rings on avatars generated by artificial intelligence

[@krishnamishra2465]

not this exactly the ring wont be the same u need a large data set to train it to perfection

[@jakemccartney1827]

when i go to test image i click the link but theres no pic that comes up its blank

[@younesdom]

there should be a .gradio.app link in the output when running this step

[@sagarjani8760]

Which model should we use from huggingface

[@younesdom]

I used 1.5 which is the default, you can try new versions tho

[@legalizenuclearboms5569]

I keep getting the message "python3: can't open file '/content/gdrive/MyDrive/sd/stable-diffusion-webui/webui.py': [Errno 2] No such file or directory" This happens every time i try using anything by TheLastBen and nobody is able to help me on it

[@SignificantOwner]

3:15 🇩🇪 Helpful video, thanks!

[@slksmg5875]

Can this model be trained on images and predict the next image

[@younesdom]

It can be trained on images and generate images according to your training data

[@ZeroXL722]

I’m having trouble with the model download part it’s gives me an error also I did all the steps you did

[@younesdom]

Sorry I do not maintain the github repo

[@thatsdoablevr]

So you can't download the trained model?

[@younesdom]

There should be a model file in your google drive that you can download

[@sk.darina]

Hi, great video! I can't download the model, it gives me HTTP ERROR 401 and I can't procced to the next step, any clues?

[@younesdom]

seems like they require a token again, Im not sure why they removed it and added it now, tbh Ive never used stable diffusion since this video, hope you will find a way

[@shalinidm8089]

did you find solution I'm facing the same error

[@psr9289]

Danke bro

[@zonealone5487]

how would i go about doing this using my own computer?

[@younesdom]

first you need a good GPU with lots of RAM, then you can just do the same thing with python if you check the code behind the notebook cells

[@zonealone5487]

@@younesdom I have a very good pc so yeah.. I just need the guide

[@twitterglobalarmy]

I utilized a ton of AI tools. Blue Willow is the one I love the most. Blue Willow is really cool, and I made some incredible images using it

[@CryptobdSchool]

Blue Willow is gaining traction for it's quality and this image generator gives anyone to create unlimited images completely free. That's really amazing!

[@sneakerhead4124]

How do you resize the images to 5.12 by 5.12

[@younesdom]

On windows you can use the fotos app but its a bit tricky, the step by step guide recommends this website to do it www.birme.net/ (tho usage without warranty from me, I did it on linux kubuntu with Gwenview)

[@sneakerhead4124]

@@younesdom what you mean usage without warranty

[@younesdom]

@@sneakerhead4124 I dont trust websites without imprints, better to do it on windows

[@kaont3569]

It didnt work because is not showing use_gradio_server. Also why are a tone of random data and images in my drive, that are not mine and not my instance images...

[@younesdom]

the python notebook generates these files

[@whellyabima]

how tostable diffusion checkpoint?

[@younesdom]

The model is being saved to your google drive, so you can use it again by clicking the checkbox as a checkpoint. Hope that helps

[@sniperplus]

How can we create an android app using it

[@younesdom]

you would need to host the model on a server (or use an existing API, which probably costs more) and send http requests from the android app to the server.. You could use something like FastAPI to create the API in python and setup the inference and training routes..

[@user-vh9sj1gw5w]

If we train our own images will it be private? I dont want my selfie online.

[@younesdom]

it will be private but it is on google servers, tho only in your google account.. If you want complete privacy you would have to run the model on your local machine which would require a good GPU..

[@user-vh9sj1gw5w]

@@younesdom Thanks man

[@SeanAlSuwaidi]

how good would that GPU have to be ?@@younesdom

[@redaizo]

How to do this localy?

[@younesdom]

To run it locally, I recommend at least a 16 GB GPU. Here is a good repo by huggingface to train your model with dreambooth: github.com/huggingface/diffusers/tree/main/examples/dreambooth

[@redaizo]

@@younesdom got two 4090, but i'm bit lost i have to install dreambooth first to SD ? thanks for the links! EDIT: found an amazing tutorial by Aitrepreneur, i'm trying this now

[@Afkmuds]

how can i train a model for free that can read pdfs and learn on that data?

[@younesdom]

first you need to collect your data, maybe perform text recognition on your pdfs and then pick a model (check huggingface and the transformer library)

[@Afkmuds]

@@younesdom yuh got the data and found a model from Facebook fast text model but it’s not reall interpreting as I’d wished, or maybe it was the trying to convert to pdfs again that messed it up idk

[@aimusiek]

Could not pass Model Download both read and write tokens from huggingface HTTPError: HTTP Error 401: Unauthorized :-(

[@younesdom]

Huggingface keeps changing things so there can be bugs... google can help

[@robcalif6337]

Your colab sheet is different than your video.

[@younesdom]

the video is pretty old and the github colab keeps updating, hope it still works

[@VioFax]

Lets just be clear, you can have a pretty decent computer and still not have enough juice to do training at home.

[@juanpaju8096]

the hugging face token space in the olab doesnt appear to me, why could tis be happening

[@rubaduba8263]

dont need it now

[@bren4425]

why so quiet??

[@younesdom]

was on vacation

[@Nelex5000]

I've noticed the dislike amount is always the same with these videos. probably real human beings that were looking for this and all never found it. interesting, isnt it?

[@MansubaTabassum-nw1kb]

This is not training a model ...it us just fine tuning

[@younesdom]

fine tuning involves training some parts of the model, so its not wrong to call it training i guess :)

[@contentfreeGPT5-py6uv]

Yea fine tunning, the title Is clickbait, but it's ok all people doing it

[@janbaer3241]

Start video at 3:00. You're welcome.

[@younesdom]

I need more watch time to Earn $ on yt :(

[@Nelex5000]

No download link: check no actual explaination on where to put the images: check very non friendly ui: check not free: check another nothingburger clickbait

[@Elba.Ginon93]

It wont work, I need step by step, AI was supposed to be easier than drawing why fo you have to be a Dr Mathematician 😢?

[@younesdom]

welcome to the life of programmerds

[@neelamverma7852]

Ima use it to make corn☠️☠️

[@younesdom]

use the force wisely

[@anunnakiiscomming9810]

>uses cloud >use your own images nope disliked and flagged for being absolutely dumb

[@younesdom]

thanks for your feedback, now go make a video about running stable diffusion on an old laptop ;) PS dont forget to subscribe

[@hpatel01terp]

I keep getting this error when im running the "test the training model" LatentDiffusion: Running in eps-prediction mode DiffusionWrapper has 859.52 M params. Failed to create model quickly; will retry using slow method. LatentDiffusion: Running in eps-prediction mode DiffusionWrapper has 859.52 M params. loading stable diffusion model: TypeError Traceback (most recent call last): File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/sd_models.py", line 338, in load_model sd_model = instantiate_from_config(sd_config.model) File "/content/gdrive/MyDrive/sd/stablediffusion/ldm/util.py", line 79, in instantiate_from_config return get_obj_from_str(config["target"])(**config.get("params", dict()))

[@JirenFTW]

Im getting the same error, it was working fine yesterday Idk whats causing this error

[@younesdom]

no idea what that is, someone please ask chatgpt..