Create AI Images Of Yourself for Free: DreamBooth & Stable Diffusion

✨ Support my work on Patreon: / allyourtech
⚔️ Join the Discord server: / discord
🧠 AllYourTech 3D Printing: / @allyourtech3dp
👾 Follow Me on X: / blovereviews
💻My Stable Diffusion PC: kit.co/AllYourTech/stable-dif...
Dreambooth URL for the tutorial: colab.research.google.com/git...
Huggingface: huggingface.co/
In this tutorial I'm going to show you how to create a model with your own images that you can use to generate an unlimited number of stable diffusion A.I. images.

[@ABQSentinel]

Just found your channel. Excellent video! I just completed the process, walking through the steps as you described and it works great! I'm looking forward to checking out more of your content.

[@allyourtechai]

Awesome to have you here!

[@macdaddymelon]

Such an amazingly helpful video, thank you!! Which video should I watch next to use the trained model I've created? I already have a local installation of stable diffusion.

[@VKTVCHANNEL]

How I missed this channel!!!!

[@AmosIntelligence]

You are probably one of the most underrated KZbinr. I cannot believe that you have so few views and subscribers haha. Thanks for your videos anyway. I am broke as f*** and trying to set my online business and most of your work is extremely helpful

[@allyourtechai]

I really appreciate the kind words, thank you! Let me know what type of content you would like to see and I’ll try to pull together some videos

[@AmosIntelligence]

@@allyourtechai I would love to but I need to finish watching your other videos beforehand. I will keep you updated but there is one thing I would love to learn: how to make 3d designs to sell jewelry on Etsy 😁

[@deltaxsingh]

Clear, step by step, foolproof. Thanks!

[@allyourtechai]

Glad it was helpful!

[@deltaxsingh]

@@allyourtechai My buddy recommended using Amazon SageMaker Studio Lab to fine-tune the model. What're your thoughts and do you think a video on the subject would be helpful?

[@uwh89]

I tried many methods. At first I ran webui locally, but my computer had all kinds of strange errors. Then I tried to run dreambooth on webui on colab, but it also failed to train. Until I saw your video and code, I finally succeeded! Thank you very much

[@allyourtechai]

That’s awesome! So happy to hear it!

[@kimchibator4457]

Well done! Thanks for the knowledge!

[@allyourtechai]

My pleasure!

[@EM-wd8wq]

This is one of the most cise video I've seen so far on this topic. Perfect.

[@allyourtechai]

Thank you so much!

[@Kelz_codes]

At last someone who actually know what they are talking about and write some cocde to run a model, subscribed!

[@yeezythabest]

This is literally the only tutorial that gave me good results thank you !!

[@allyourtechai]

Glad it helped!

[@yeezythabest]

@@allyourtechai do you have a tutorial or infos on how to use the model trained with this notebook but using colab ? With automatic 1111. I saw your other video but it's for local only (I think)

[@allyourtechai]

@@yeezythabest I'll see if I can pull togehter a tutorial for that :)

[@mikejjOZ]

Thanks great tutorial! What kind of model are you creating here though? I mean how do you use it outside of dreambooth?

[@Stick3x]

Mind blowing!

[@user-sw9pw3lq8s]

Thanks for your video. It is very helpful! If I could like to train another dataset, what steps from the video should I repeat ? I assume that the first step, installation could be skipped right ? Could I use the same google account for training of different datasets ? Thanks !

[@stinkyjutsu]

I spent weeks and literally thousands of dollars to try and get dreambooth to work, and then turned around and trained a character through a hypernetwork in under an hour with excellent results. But I'm not giving up on dreambooth just yet.

[@yeezythabest]

Which hypernetwork did you use?

[@makulVR]

Is there a 2nd part to this video? This was absolutely awesome!

[@allyourtechai]

Thank you! I've done a few follow ups, but would love to hear if you would like me to cover anything specific?

[@astonteague8260]

@@allyourtechai what are the specific videos? I am on a Mac. Can I still use this model?

[@nerdygirl430]

Great video ! Very useful ! I created high quality images. Could you (1) create a video about how to continue running after getting disconnected from the colab (2) And how to run the saved ckpt model at different google account ? Thanks !☺

[@jakedoesgaming9373]

on the model card on the hugging face webiste me and a few other memebrs of the community cannot find the box that you have to tick for the process to work can i get help on this ?

[@DanielPaast]

have followed your instructions. the checkpoint file in the relevant folder. however, SD creates only identical photos as the learning photos.

[@naqeerezaie]

Thanks for great content, I just have one question that how to bring this session again and how to load our trained model once reopen Google-Colab?

[@allyourtechai]

What I normally do is train the model in colab, then load it into a local copy of InvokeAI or automatic1111 on my home pc.

[@nekineli]

Amazing presentation. But I have a question. If I understood properly, this is using an Google GPU to train the model. But I found in another videos from about 5 months ago that we needed to have a high end GPU. What is the difference between the two situation ? Thanks !

[@allyourtechai]

Great question. You use a GPU on googles cloud to train the model, but when you run it locally on your home PC, you will need a relatively high end GPU to actually create images using automatic1111 or invokeai software.

[@nekineli]

​@@allyourtechai Thanks for the answer ! I eventually figured it out and now I'm experiencing dreambooth via runpod

[@surawatloetsivanon8099]

Your video is awesome. By the way, I have to train many people's face to only one model. Do you have any idea?

[@HSR]

Need a video like this one on how to reuse that saved model.

[@allyourtechai]

I’m on it!!

[@maddoxmakesart]

@@allyourtechai Which would be the next Video! :D I started following 2 days ago - I have already made it this FAR! LFG!

[@LimousineIndianapol1]

I am now working on training a text model that works. I think there must be a translation interface between the words and the model. The interface will be the params for the actual text. Example: 1. Enter your text: Hello World, 2. What Font, 3. Font Style, 4. Font Weight, 5. Font Color,....until you have all parameters for the text. This would allow for text formatting instead of it using images to create text. How does ChatGBT create text without it being a mess?

[@dhivimhotep]

Hello, when I train a LoRa model each time the result is a file safetensors of a size 37.9 MB. I see other LoRa models with 75.6 MB or 151MB! how to increase this size up to 151MB? what are the parameters involved?

[@LH002_]

How to save the predefined colab? I've trained everything and stuff, but it seems I have to do it again if I close the tab

[@ChrisCoelho]

when i'm about to run the conversion step i get this error "NameError Traceback (most recent call last) in () 1 #@markdown Run conversion. ----> 2 ckpt_path = WEIGHTS_DIR + "/model.ckpt" 3 4 half_arg = "" 5 #@markdown Whether to convert to fp16, takes half the space (2GB). NameError: name 'WEIGHTS_DIR' is not defined

[@bongpng]

first of all, amazing video! I am having a problem, I used all my GPU available, and when I try to run it locally, following the steps at the "train your model" tab, it does not compute, it takes 8s to run the cel and gives me this message at the end after completing: AttributeError: 'NoneType' object has no attribute 'cuDeviceGetCount'

[@worawut-4233]

01:30 how to fix? 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. torchaudio 2.0.2+cu118 requires torch==2.0.1, but you have torch 2.0.0 which is incompatible. torchdata 0.6.1 requires torch==2.0.1, but you have torch 2.0.0 which is incompatible. torchtext 0.15.2 requires torch==2.0.1, but you have torch 2.0.0 which is incompatible. torchvision 0.15.2+cu118 requires torch==2.0.1, but you have torch 2.0.0 which is incompatible.

[@LanceJordan]

Where are the follow-up videos for this? thanks.

[@philipeaugustoleao9138]

how do i do to use the same model again without going to the same step by step after the first use?

[@miwei5513]

Is there an updated version available, since I am getting errors during step 2 (Install Requirements): Building wheel for ffmpy (setup.py) ... done 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. llmx 0.0.15a0 requires cohere, which is not installed. llmx 0.0.15a0 requires openai, which is not installed. llmx 0.0.15a0 requires tiktoken, which is not installed. tensorflow-probability 0.22.0 requires typing-extensions

[@MathiasRehfeldt]

i am haveing the same issue

[@LintBisquick]

@@MathiasRehfeldt Any progress on resolving this?

[@MathiasRehfeldt]

unfotunately not :( @@LintBisquick

[@LintBisquick]

Yea it sucks. Really want this to work. Ive seen Cuda fixes but this is not my skill set. Need exact instructions. If you find any other avenues to achieve the same thing I'd appreciate the same info. I will reply here if I find a fix or another way. @@MathiasRehfeldt

[@MathiasRehfeldt]

I only have found a paid solution. I am using astria.ai to build my model and then it is possible to use astria with some limitations to render images. Another possibility is to upload your model to rundiffusion.com this is basically stablediffusion online. Its payed by the minute though @@LintBisquick

[@AndySovaga]

Helpful video, but I can't download the model file to my googledrive, because I don't find the licence accept button on the hugginface model page. :(

[@FelixTheAnimator]

Ok that first step. What are they training the model with? Where is it getting that?

[@allyourtechai]

I took photos of myself and uploaded them to train the model. YOu can provide any images you wish.

[@user-to8tn4gf3m]

I cant find the check box on the page. Where is it? On the huggingface page? Where? i am checking and checking

[@theterminaldave]

I think I've asked in other videos of yours, but if my computer is a potato (low ram, older cpu, 1060 3g), will i be able to run this?

[@allyourtechai]

No, unfortunately you need at least 8gb of vram and a more recent gpu to get decent performance and in many cases to get it to run at all

[@theterminaldave]

@@allyourtechai Again, I appreciate your response!

[@hadihonarvarnazari4941]

How to generate image based on a given image? I want to avoid changing the scene and only want to change the style.

[@arafatnehan9402]

Im getting the following error when running webui.bat :( can you please help? ERROR: Could not find a version that satisfies the requirement torch==1.13.1+cu117 (from versions: none) ERROR: No matching distribution found for torch==1.13.1+cu117

[@MattHanr]

Install those missing packages, for example pip install torch or pip3 install torch

[@Azzalto40]

and about stable difusion??

[@theCreativeAddict3d]

Thanks for the video. I waited for around 20 minutes, after making the changes, but it is stuck and showing "WARNING the following values are not passed to _________________ (and shows a blank)" Should I wait more?

[@allyourtechai]

If you hit the X next to the error it will clear and show you the progress. Expect that step to take about an hour

[@theCreativeAddict3d]

@@allyourtechai Thank you so much Sir :)

[@jasplioin]

great video, i created convincing image of myself on colab, but when i import the exported model into automatic1111 and run locally, the images are totally different, do you know what could be the possible reason?

[@allyourtechai]

Could be the sampler you are using, the prompt, number of steps, etc. it takes some tuning and tweaking to get great quality. I’ll do a video about it

[@darksidekurd]

What is the name you have entered as model name? It is important that you write the correct model name in the prompt. I, for example, entered just my first name but not my last name and I also got totally different results. Using the correct model name resolved the issue for me.

[@abdulmutaalazhar8235]

@allyourtechai Some of my images are high quality, so 512 x 512 is really small part of the image (Probably covers only an eye). How can I use those pictures for my training data ?

[@allyourtechai]

The higher the resolution, the more powerful the hardware needed to train the model. On an RTX3090 for example, 1024x1024 is about the max resolution you can train. It will be easier for you to train smaller images and upscale.

[@Ausomapto1]

Hi, hope your'e doing good! THe thing is when I'm doing the training part I'm getting a message saying "PyTorch has CUDA Version=11.7 and torchvision has CUDA Version=11.8. Please reinstall the torchvision that matches your PyTorch install." Can you please tell me what do I do to fix this issue?

[@allyourtechai]

That’s a new error to me. Let me see what I can find out

[@HeenaSheikh]

Facing same error. Please help

[@elishevas6302]

Hi. Trying to gollow your video but keep getting: python3: can't open file '/content/train_dreambooth.py': [Errno 2] No such file or directory. What Did I do wrong?

[@bhavanakedari1009]

were you able to fix the error? can you please help me out?

[@JoqoAlfredo]

Great video! However I am getting the following error and don't know how to fix it: HFValidationError: Repo id must use alphanumeric chars or '-', '_', '.', '--' and '..' are forbidden, '-' and '.' cannot start or end the name, max length is 96: ''. I would appreciate some help, thanks!

[@allyourtechai]

do you have really long file names or files with special characters in them?

[@jensenjacob7810]

does this still work? It doesnt for me, idk if its something I did wrong but if anyone knows whether this still works please leave a reply, thanks!

[@user-sw9pw3lq8s]

Is it possible to use a photo that has more than 1 person ?thanks.

[@allyourtechai]

Great question. I will test and find out!

[@harristamsi6590]

This don't work anymore. I made some of the models using this dreambooth link. Is there updated tutorial video for this?

[@allyourtechai]

I have several updated videos on the subject

[@twltch]

Can you export this training model to use in Stable/automatic1111?

[@allyourtechai]

Definitely! Just copy the checkpoint into your local automatic1111 models folder and restart. Should show up in your list of models.

[@kk47shooter91]

i just started getting this error today. 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. torchaudio 2.0.2+cu118 requires torch==2.0.1, but you have torch 2.0.0 which is incompatible. torchdata 0.6.1 requires torch==2.0.1, but you have torch 2.0.0 which is incompatible. torchtext 0.15.2 requires torch==2.0.1, but you have torch 2.0.0 which is incompatible. torchvision 0.15.2+cu118 requires torch==2.0.1, but you have torch 2.0.0 which is incompatible.

[@allyourtechai]

Interesting, let me see if I can find a workaround

[@kk47shooter91]

@@allyourtechai please ives been pulling my hair out for two days.. just started happing out of nowhere on my colab

[@daraneepavin]

I tried training and put 200 pictures but the face came out weird please help

[@Rasel720]

Thanks . I have been able to create my own avatar following your tutorial. but my face was disaster in the rendered pics. i think low size of the photo is the reason behind it. my photo was 60kb avg after cropping from online. (i subscribed your channel after this detailed , clear tutorial)

[@allyourtechai]

Getting a really crisp image of your face even in the 512x512 images is really helpful in training the model. Let me know if I can help with anything, i'm happy to have you here :)

[@juancarlozamora5285]

Hi! Did you get to fix your face? I'm having the same problems, I don't know how I can start over.

[@Rasel720]

@@juancarlozamora5285 yes i fixed it. my previous photo was long ranged. so the Ai was struggling to get the face details. After uploading the short ranged pics where face was close, the problem got solved

[@krizh289]

Did this locally on my laptop's rtx3060 and it took 3 hours to train my face...this should be much faster lmao

[@darysi6338]

What do you do for a living outside of making these great youtube videos just curious~

[@allyourtechai]

Thank you! My day job is VP of Product for an ad tech company.

[@alexsanders8881]

hello friend, is it possible to use models already trained for this process?

[@allyourtechai]

Yes, you can find checkpoint files all over the place that are trained on different data sets. So many options!

[@alexsanders8881]

@@allyourtechai Right, but in this link provided by you where I put the "link" of this model I want to use?

[@philipeaugustoleao9138]

@@alexsanders8881 I have the same question

[@illxet]

so what is the process if we don't want to use online services?

[@allyourtechai]

I believe you can train a model on automatic1111, but I haven't explored that personally since dreambooth works well.

[@mathwiz1260]

hey bud , I'm having an error in the last line of install requirements; 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. llmx 0.0.15a0 requires cohere, which is not installed. llmx 0.0.15a0 requires openai, which is not installed. llmx 0.0.15a0 requires tiktoken, which is not installed. tensorflow-probability 0.22.0 requires typing-extensions

[@perlyngemark2927]

Same problem here, exactly the same

[@LintBisquick]

ME 2. Any progress resolving this issue?@@perlyngemark2927

[@LeeXuan88]

Ive been searching this kind of technology so that i can upload/use it for my instagram profile. As I'm too lazy to take the real photo or hiring a professional photographer. As the same time I want it to look real me, no cartoon or art. Completely like the real me, same height, white, skin color, same face accurately, only the output look better like the photo been taking by profesionnal photo grapher. Do you think this would work? Is there any paid platform that can achieve this instantly? Thanks

[@alejocorreagomz]

app lensa

[@LeeXuan88]

@@alejocorreagomz tested.. the output suck,, too cartoonist.. and every time u want to generate it asked for the money.

[@ah-ry8em]

Mine is taking forever to upload the pics, like over 10 minutes now for ~70 pics. Is this normal?

[@allyourtechai]

Depends on the speed of your internet generally, but I wouldn’t expect it to take more than a few seconds in most cases

[@astonteague8260]

I got an error when I tried to generate the 4 samples. Not sure why. Any ideas?

[@allyourtechai]

Do you know what the error message was by chance?

[@popgun427]

I don't see the box 😢

[@jack_dawson]

Which video of yours is initial point ? What video is first video for ai?

[@allyourtechai]

What do you need help with?

[@jack_dawson]

@@allyourtechai initial installations for generating images.

[@jack_dawson]

@@allyourtechai this video include every installation and plug in.?

[@wherewolf87]

How did you crop to 512x512??

[@allyourtechai]

I used adobe express to do it, but any photo editing software will allow you to crop the images.

[@1996sagark]

new huggingface model card doesnt have any tick button. Terms are accepted by default for new account. Thank me now!!

[@allyourtechai]

Haha, thank you!

[@urgoku]

@@allyourtechai dear, while running the acceleration Launch Train, I gave a long timeout… it got interrupted. Do I have to start from the beginning? Will the same token work ?

[@1manfactory]

Too bad. This does not work anymore. Runs into errors even during installation.

[@allyourtechai]

I haven’t been able to find a new working September colab. I’ll see if I can put one together this week

[@androidgamerxc]

im doing this right now and will update you if it works

[@allyourtechai]

Let me know how it goes!

[@androidgamerxc]

@@allyourtechai it finihsed in 47 min which i think was quite some time i have 3080 and i put it in 1111 and for me since i am not that great at promt so what i do is copy paste other people prompts and the images are absoulty awesome thank you so much for this and please can you make video about using realistic1.4 model and making lora model with it sorry my english is bad

[@allyourtechai]

SO glad to hear that it worked for you! I will do some more videos as you suggested :)

[@androidgamerxc]

@@allyourtechai also make a discrord for your channel where people who learned thing or two from your channel could post art work there and tell prompts etc

[@allyourtechai]

Join me, and a growing community of enthusiasts on the all-new All Your Tech Discord server: discord.gg/bhNj7xyjkA

[@viper_eyes]

isn't 3000 steps is over trained model ? doesn't this loses the styling options ?

[@allyourtechai]

It depends on the number of images. I've heard 10 steps per image, but I may play around with it more and see how much of a difference it makes.

[@steveherman1395]

@@allyourtechai I hate asking a question when I don't know what I am asking, but 10 steps per image X 30 images is 300?

[@anovin82]

The number of optimized training pictures to use is definitely 20. Any more than that will actually not improve your model.

[@FelixTheAnimator]

I don't want to train my model on somebody else's images (even if they say I can) 'cause I know dang well that casual art theft is a thing. Maybe I'm just a control freak...

[@allyourtechai]

You can use any set of images you chose. It's up to you what you provide the model to be trained with.

[@castielvargastv7931]

the following vignette you use for your videos looks strange😁😁

[@allyourtechai]

I turned it off for all of my recent videos. Apologies if it was distracting

[@azadarhussain4808]

love from pakistan

[@sheeesh404]

Doesn't work anymore :\

[@allyourtechai]

planning to do an update and also a local tutorial for those with a powerful gpu.

[@Famexcess]

Would greatly appreciate an update for the google colab method@@allyourtechai

[@gu9838]

great for people with huge egos lol j/k ;-)

[@allyourtechai]

Lol!

[@MakoraPhotobooth]

louder please

[@aviadd843]

WTF... Why do we need to install this crazy procedure? Can't it be done online? Just generate from pictures I load online, is that so hard? Why do I need to install 6GBs? Nooo, don't think I'm gonna do this.

[@khalid9313]

Can't get my face right. Weird face

[@allyourtechai]

You need to play around with the prompt, sampler, steps, etc. also be sure to use the face fix feature.

[@zeezus890]

@@allyourtechai how do you use that feature?

[@Meshulator]

outdated, broken

[@allyourtechai]

I make a new one based on the changes in recent months. Thanks for the heads up

[@wulfgangaudio]

I want to execute "run to generate a grid of preview images from the last saved weights." within DreamBooth_stable_Diffusion.ipynb But this code doesn't work: (can you please help me!) (I am not a coder, so please don't slap me for this!) #@markdown Run to generate a grid of preview images from the last saved weights. import os import matplotlib.pyplot as plt import matplotlib.image as mpimg weights_folder = OUTPUT_DIR folders = sorted([f for f in os.listdir(weights_folder) if f != "0"], key=lambda x: int('5')) row = len(folders) col = len(os.listdir(os.path.join(weights_folder, folders[0], "samples"))) scale = 4 fig, axes = plt.subplots(row, col, figsize=(col*scale, row*scale), gridspec_kw={'hspace': 0, 'wspace': 0}) for i, folder in enumerate(folders): folder_path = os.path.join(weights_folder, folder) image_folder = os.path.join(folder_path, "samples") images = [f for f in os.listdir(image_folder)] for j, image in enumerate(images): if row == 1: currAxes = axes[j] else: currAxes = axes[i, j] if i == 0: currAxes.set_title(f"Image {j}") if j == 0: currAxes.text(-0.1, 0.5, folder, rotation=0, va='center', ha='center', transform=currAxes.transAxes) image_path = os.path.join(image_folder, image) img = mpimg.imread(image_path) currAxes.imshow(img, cmap='gray') currAxes.axis('off') plt.tight_layout() plt.savefig('grid.png', dpi=72) --------------------------------------------------------------------------- ValueError Traceback (most recent call last) in 10 col = len(os.listdir(os.path.join(weights_folder, folders[0], "samples"))) 11 scale = 4 ---> 12 fig, axes = plt.subplots(row, col, figsize=(col*scale, row*scale), gridspec_kw={'hspace': 0, 'wspace': 0}) 13 14 for i, folder in enumerate(folders):

[@RichardLucas]

Doesn't work, any more. Lol, I searched Google high and low and found many, many others reporting the same thing: Dreambooth + Automatic1111 just stopped working as it should, and no plausible explanations have been found.

[@aalore7020]

A very practical operation tutorial, thank you very much for your dedication. Through your video, I quickly got started with DreamDance, but now I have a new problem. I want a photo similar to the one below（i.imgur.com/OJbkF3t.jpg）, but I hope to change the cat's face in the picture to my own cat's face through stable diffusion. I don't have any ideas for achieving this goal. Do you have any ideas? Thank you very much for your help!

[@daviv]

how do i use the model that is saved on google drive again after i closed the collab?