Dreambooth in Automatic1111. Cpu only & gpu option.

[@valscripted]

thank you for being up to date so fast and creating tutorials that are, not only nice to follow through, but quite funny as well!

[@sebastiankamph]

Glad you like them! Thank you for the kind words :)

[@amj2048]

IMPORTANT FIX FOR PEOPLE GETTING ISSUES: (or at least one fix) Set "Save a checkpoint every N steps, 0 to disable" to a value that will never be reached, so set it to something like 5000 Also set "Generate a preview image every N steps, 0 to disable" to a value that will never be reached. I just set both of those to 5000 (so they would never be reached) and I can successfully train models locally now. If you set those settings to 0, or a value that can be reached like the default 500 (which I did before) you will get an error. (I have an RTX 3060 with 12GB of vram)

[@sebastiankamph]

Very interesting, and a good tip!

[@McMartinLC]

Thank you! Have been having that exact error "Attempting to unscale FP16 gradients" too! Nice work and thanks again for sharing

[@amj2048]

@@McMartinLC the credit actually goes to Aitrepreneur who discovered this fix 🙂

[@velenagm]

The dad joke about installing kitchens was just the icing on the cake! Unexpected and therefore hilarious! Thank you for the up-to-date content btw. Your videos are very helpful

[@sebastiankamph]

Glad you enjoyed it! And thank you for taking the time to leave such a kind comment 😊

[@anonymousmuskox1893]

If you are getting errors make sure the entire folder path that has stable-diffusion-webui has NO spaces in it.

[@blimeyitsRichard]

aha :)

[@DeliciousBrain12]

ohhh ok thx for that! And if I got spaces in it...do I have to de- and reinstall everything to rename the folders?

[@amj2048]

oh crap, that would probably be my issue too, I'm going to have to move some things around

[@blimeyitsRichard]

@@amj2048 just take the spaces out.. everything still hung together for me

[@blimeyitsRichard]

@@DeliciousBrain12 just take the spaces out.. everything still hung together for me

[@DrewMedina]

Just started using this in Visions of Chaos, so awesome! (Front end gui app). Thanks!

[@violentpixelation5486]

You being able to use "Visions of Chaos" means I want to have your PC setup.

[@sebastiankamph]

I've been playing around with that, it seems very powerful from what I've seen!

[@sommie666]

Best tutorials out there!

[@violentpixelation5486]

Thanks alot for the walkthrough. Running on a RTX2060 over here so I'm very pleased with the newly available "cpu option". I'm not a native speaker, but the "counter-fitting" joke won me over. :D

[@sebastiankamph]

Haha, glad you liked it 😊

[@myesterious]

run on cpu is how many time more than gpu ?

[@Amarand]

Fantastic tutorial! Would love to see more of this!

[@sebastiankamph]

Glad you like it! Got a whole bunch of tutorials on the channel if you want to see more :)

[@hexcode5703]

Whata amazing update!

[@sebastiankamph]

Right!? From complicated installs to this. This is a real gamechanger.

[@maorsabag828]

Amazing tutorial, quick and clearl to understand, thank you very much!

[@lavahawk]

amazing resource, thank you for showing the charts any everything!

[@CoconutPete]

this is pretty wild and going to have to give it a try... the only part I'm confused about is the model you use with a trained model. Is the model you create one that works on its own or do you select the default sd model or any other model you wish?

[@6EXAI]

i installed dream booth and restated it says i installed it but i don't see the tab any suggests?

[@testales]

Argh! I just gave in this morning and bought a used RTX3090 to finally get Dreambooth running and now it works out of the box just by installing an extension, even in CPU mode?! I'll try this ASAP, so far I only got all sorts of CUDA and VRAM related errors with my good old GTX1080 TI after following complicated guides.

[@sebastiankamph]

Your purchase of 3090 is not a mistake! CPU is suuuuuuuuuuuuuper slow. 3090 godmode 🌟

[@testales]

@@sebastiankamph Still getting CUDA out of memory errors. When I try to use fp16, I also get an error "expected scalar type Half but found Float" - any ideas how to fix that? According to your sheet it should save a little more VRAM.

[@sorijin]

Homie how much you get that 3090 for and from where? Im about to buy a new desktop gpu

[@magistrcooldayn233]

Nvidia stopped selling 3090 officially, you can still find what left on stock or buy used one

[@sebastiankamph]

@@testales I saw in github about that scalar half thing getting fixed just hours ago. Try updating.

[@newbiecryptewbie3170]

Getting error "RuntimeError: Detected that PyTorch and torchvision were compiled with different CUDA versions. PyTorch has CUDA Version=11.6 and torchvision has CUDA Version=11.3. Please reinstall the torchvision that matches your PyTorch install." Any recommendations?

[@campbellspeelman3491]

Hey guys - copied from another comment but if you restart AUTOMATIC1111 and there is no Dreambooth tab, navigate to where you have your "stable-diffusion-webui" folder and add "accelerate" to a new line at the bottom of the "requirements.txt" txt document, save and start automatic1111 again

[@anonymousmuskox1893]

i'm on ubuntu and this worked for me. thank you!

[@JUUSOPILOTTI]

my stable-diffusion-webui folder is empty 😮

[@JayFiggyy]

dreambooth not loading in my stable diffusion?

[@devnull_]

Thanks. I really hate these installers that are doing "something" without showing any progress bar.

[@mercyveritas1125]

How long did it take u to train in ur case?

[@thomas_w]

Really cool video, what prompts did you use for these pictures? I tried getting a gta v style charavter of myself but havent really found a good promt yet :/

[@sebastiankamph]

Thank you! Try this: kzbin.info/www/bejne/sJKreZ-Zj8SgpJo

[@fashionedsoul8935]

Hello! I recently installed stable diffusion using A111 using one of your guide videos and everything worked great! Though I tried to install Dreambooth and get nothing but error after error when trying to use it. Was wondering if you have a more up to date guide for installing and using it with A1111.

[@Snafu2346]

I haven't updated, mostly because everything on auto1111 has been working fine. I still use google colab to create a model.ckpt. Mostly because it seems faster than if I do it on my own computer. I have a RTX 3080, but using the in house version of ckpt seems slower than using the google colab and then just saving it and downloading the ckpt. I know if others have better success.

[@sebastiankamph]

This is my preferred choice as well. It's the most reliable and in that sense, fastest.

[@swannschilling474]

Thanks for spreading the hot gossip!! 🤩🤩🤩

[@sebastiankamph]

You're most welcome! I'll make sure you always have the latest gossip.

[@mattc3510]

Thank you for showing us a tutorial with all settings for us to follow along with huge help. 👍

[@sebastiankamph]

You're very welcome!

[@AndyHTu]

If you picked an existing model, you add training to it, the trained material only exists in that model correct? Or do you have to save a fresh new model?

[@sebastiankamph]

You're adding training to an existing model, yes. Then that new model is saved as a fresh model. The previous model is left as is.

[@FLIPPS]

I trained my model the way you described, but now every image that Stable Diffusion creates looks exactly like the input images, including the background. The rest of my prompt is completely ignored, whether i write "XY in the style of (artist)" or "XY swimming in a pool" or something else. Why does the newly trained word overwrite everything?

[@sebastiankamph]

Then the model has been overtrained. Try setting XY in the end of your prompt.

[@FLIPPS]

@@sebastiankamph Thank you, but I still dont quite understand. So the new CKPT-file should be used, right? When I do this, EVERY prompt is XY, even if XY is not part of the prompt text. For instance the prompt "cat" still ends up being a picture of XY. I redid the training with a completely different picture set and the problem is the same. There seems to be nothing in the CKPT-file but the new XY from the training images, so "cat", "car" or "landscape" always is XY.

[@sebastiankamph]

Then you need to lower training steps

[@morozov_creator]

Did you solve the problem? The same situation. 11 incoming photos, 1100 steps, 0.000005 learning rate. Everything I enter is the same as the original photos.

[@FLIPPS]

@@morozov_creator I reduced the steps to ~20 per image. Now it doesn't overwhelm the rest of the prompt, but there are also way less similarities with the trained persons face

[@christoffermonikander2200]

Automatic Center Crop. I think I finally figured out the reason why Stable Diffusion creates so many images that are cropped.

[@tobyseaman7977]

Looks awesome! How did you manage to work around the lack of xformers on Windows?

[@tobyseaman7977]

resolved: use the recommended version of Python!

[@sebastiankamph]

Glad you got it resolved!

[@LoneRanger.801]

Haha.. loved the counter fitting joke 😂😁

[@sebastiankamph]

Haha, glad that got a laugh somewhere 😉

[@PedroHenrique-xm1yt]

Thanks for the video! I have a problem though, i'm using a 6gb 1060 which i don't expect to work, so i was going to try using my cpu. However, when i enter the model name and select a checkpoint, and click create, 10 seconds later i get a CUDA error (ran out of memory). Do you have any idea why that happens? i'm not even training anything yet, i can't even get to the part where i select training with cpu lol

[@kimhan2615]

Got the same problem and now it has many new features in the new updated version. it seems that it needs VRAM to extract the model.

[@PedroHenrique-xm1yt]

@@kimhan2615 welp, that sucks. Did you ever get it working? I don't know why, but i feel like it will probably never work for me lol

[@devnull_]

How about having xformers enabled? Yes or No?

[@amj2048]

Thanks for sharing. I just tried this and I'm getting lots of memory issues. (I have an RTX 3060 12GB). I tried using 8-bit Adam, which did help get a bit further, but still had memory issues. Then I tried using 256x256 images and that got to half way (from 1000 training steps), but then I got an error about "Attempting to unscale FP16 gradients" and I have no clue what that is about, so I've stopped for now. It feels like I'm close to getting this working though, I'll try again tomorrow.

[@ac3d657]

nope.

[@amj2048]

@@ac3d657 lol

[@samanmonazami9156]

@@amj2048 I have 3060 too, let us know how it worked for you thanks

[@magistrcooldayn233]

If 12 gb is not enough, how muck vram it needs? 20?

[@amj2048]

@@samanmonazami9156 I've having issues with it, it doesn't seem to be a memory issue though, I think I've got an issue with my setup.

[@ithanhunt3250]

*Can stable diffusion be installed on a 32-bit pc and only cpu ?*

[@N0XQS]

If you install under Linux and there is no Dreambooth tab after you installed the extension, add accelerate to the requirements.txt and restart.

[@6EXAI]

how do i do that?

[@N0XQS]

​@@6EXAI you open a terminal window, then you cd ~/stable-diffusion-webui and in there you edit the requirements.txt file with for example nano requirements.txt. Then at the bottom you add a new line and type accelerate Save it, close it and restart stable diffusion.

[@campbellspeelman3491]

THANK YOU

[@neisan92]

@@N0XQS Um this did not work for me, "accelerate" was already there near the top. I deleted that and placed it at the bottom, nothing, kept it in both locations, nothing. Left it in the same spot, still no tab popping up.

[@Outthebots]

what prompts for those cool pics?

[@sebastiankamph]

Different for different pics. I'd be happy to help you if you come to our Discord (see channel description) and ask there more specifically what image you want.

[@VIpown3d]

Jätte bra igen!

[@sebastiankamph]

Your Swedish is excellent! 😊

[@plagiats]

Now crossing my fingers for DreamFusion to be next!

[@sebastiankamph]

Crossing your 7 fingers on one hand SD-style?

[@lavahawk]

its out now!

[@yanwittmann]

I was successfully able to launch the training using the settings you showed in the video in the latest version of the extension, but the resulting checkpoint does not know any of the information from the source images. I used 10 images of myself with 1100 steps and the default learning rate, "firstnamelastname" as instance prompt and "man" as class prompt. I tried it with/without classification images (150) and also with 4000 steps. Nothing seems to work, I can't get the model to produce images like the source image. Does anyone know a solution for this?

[@sebastiankamph]

You can try raising your weight in the prompt with (((firstnamelastname))). If it still doesn't give you anything, I'd try retraining with more steps. Usually, people have been requiring less training steps with this version, but your mileage may vary.

[@yanwittmann]

@@sebastiankamph thanks for the quick answer, will try that tomorrow!

[@nic-ori]

Thanks.

[@tigerxplso]

If i have 8 gb of VRAM on my laptop, can i still run it but it will be slower, because it will have to dump the memory to ram/disk and retake it later when needed, or will it crash at like 80% of work done and i will just waste my time? :D

[@romulo4804]

I also want to know

[@sebastiankamph]

It will sadly not start in gpu mode until more optimizations are made in development.

[@tioedu_]

now it seems that it doesn't work anymore, in the new version

[@gumvue.studio]

is it possible to train at once 10 different people? or I need to do this 10 times? like am trying to get a Movie characters , so I have 10 of them with names

[@professor-seba]

Hey there, thanks! I am refreshing but I can't see any EXTENSION tab on my Stable :(

[@GarethOwenFilmGOwen]

Update SF

[@sebastiankamph]

Then you need to upgrade to latest version. Go into CMD (command prompt) to your SD folder. Then type: git pull Restart and try again.

[@GarethOwenFilmGOwen]

@@sebastiankamph It worked when i updated in in the CMD

[@professor-seba]

@@sebastiankamph super grateful! thanks! I always execute the git pull... apparently I had another problem with GIT or something so I had to uninstall everything and start from scratch! Thanks! :D

[@bobdrozd6680]

@@sebastiankamph same problem. After git pull *Already up to date.*. I use windows 10. do you know how can I fix dreambooth tab?

[@FullSailGameDev]

For those who are wondering how long CPU takes? I'm sure it varies by CPU. I have a AMD Ryzen 9 3900X 12 core 3.8ghz working on 3000 steps takes about 20 hours to complete. Anyone else try CPU? How long did it take and what model CPU?

[@sebastiankamph]

Great info, Johnathan! Thanks for sharing 🌟

[@morphman86]

I'll come back about that at some point after I've tried with my Quad-Core i5 3.7GHz. Might be on the discord though, because I doubt I'll remember to get back here. I'm ok with it taking time, if it works. Rather slow and working than failing super fast.

[@Vincent-mx4rk]

Hey Sebastian, do you recommend Stable Diffusion 1.4 or 1.5?

[@sebastiankamph]

In/outpainting is a little better in 1.5, but honestly, apart from that I'm not super impressed in 1.5 compared to 1.4.

[@patrickmros]

I'm getting "expected scalar type Half but found Float" every time i try to get it working on my 3060 with 12 GB RAM. Someone suggested to use "--no-half" as a start parameter but no luck with that either. I got it working under wsl but it won't work with windows so far...

[@SuperFunHappyTobi]

Same issue on my end

[@patrickmros]

@@SuperFunHappyTobi I solved it by manually deleting the dreambooth folder in the extensions folder. It worked afterwards. Well not exactly. I had to set the preview image generation to some high value because there's a bug and someone had to do the same with the checkpoint. But after this it worked and I successfully trained some models after that.

[@sebastiankamph]

Great to hear it worked for you in the end!

[@FlameWinds]

Have this error: "RuntimeError: CUDA out of memory. Tried to allocate 26.00 MiB (GPU 0; 12.00 GiB total capacity; 11.20 GiB already allocated; 0 bytes free; 11.31 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF"

[@FlameWinds]

Need in models\dreambooth\YOUR_NET_NAME\db_config.json - change mixed_precision": "fp16" to "Accelerate()"

[@ai.rtistic.whatif4691]

@@FlameWinds didn't work for me i keept getting denom = (exp_avg_sq.sqrt() / bias_correction2_sqrt).add_(eps) RuntimeError: CUDA out of memory. Tried to allocate 114.00 MiB (GPU 0; 16.00 GiB total capacity; 15.00 GiB already allocated; 0 bytes free; 15.18 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

[@madebyrasa]

nice, i just noticed this too. going to fire it up. only 2080ti, so 11g, fingers crossed.

[@stefanb7015]

Shout if it works, im on same (ti)...:P

[@madebyrasa]

@@stefanb7015 yup no go, tried everything in the ui. restarted a1111 after every attempt. Even cpu fails. I love the idea, so watching bugs on git and testing when I have time.

[@stefanb7015]

@@madebyrasa I'll get back to you in here when I've tried..

[@stefanb7015]

@@madebyrasa I get the same, tried everything. Also failed with the cpu (I'm on i9 and 32 RAM). It seems to still be a mess, but it will be sorted out in a couple of weeks. :P

[@Jack1998d1]

@@madebyrasa works for me on 2080 ti, are you getting errors in your console? I'm on Linux so maybe it's a windows thing?

[@tylerd3306]

so in essence, this just adds whatever you train to the stable diffusion model, which is triggered by certain words? Technically have access to the original model with an addition?

[@sebastiankamph]

Yes! You're training the model to add your input images, which could be for example a person or an art style.

[@blackvx]

It wasn't working at all for me. My machine has all the requirements to run Dreambooth (RTX 3080 12GB). I finally found the issue! there was a warning message every time I started the webui-user : [!] Not using xformers memory efficient attention. I just needed to add xformers in the file: set COMMANDLINE_ARGS= --xformers Now it is working and my GPU is at 97% on the job 😁 .....and done! It works!

[@sebastiankamph]

Very nice! Happy you got it working and thank you for sharing what helped ⭐

[@colinmillar9804]

Thanks heaps for this was trying to figure that one out

[@anonymousmuskox1893]

add it to what file?

[@sebastiankamph]

@@anonymousmuskox1893 webui-user.bat

[@tobyseaman7977]

@@sebastiankamph Any Idea what version of python you have - this xformers is really causing me issues!

[@lolamoorehead9303]

Im getting a runtime error: expected scaler type is float vs weight when i try to train. any idea?

[@yatobatoshu8299]

same problem here

[@lolamoorehead9303]

@@yatobatoshu8299 ill update if i find a fix, so far its random. im running on a 3090 too so it isnt gpu

[@lolamoorehead9303]

@@yatobatoshu8299 multiple uninstalls and reinstalls. dont change a single setting and run it, sometimes works sometimes doesnt but im thinking will need to wait for update on it.

[@sebastiankamph]

Someone mentioned setting the preview values higher than the training steps. Apparently some kind of bug. Might not work for you though but did for many others.

[@simonavittoria586]

Hello :). I became an error in webui after installation. Error running install py for extension sd_dreambooth_extension. Command: "...\stable-diffusion-webui\venv\Scripts\python exe" "extensions\sd_dreambooth_extension\install py" Error code: 1 stdout: Installing Requirements for Dreambooth. stderr: Traceback (most recent call last): RuntimeError: Couldn't install Dreambooth requirements..stdout:

[@sebastiankamph]

Hey! Hmmm, haven't seen that error before. Make sure you have Python 3.10.x installed and if you do, try reinstalling again. Heard people having issues but most have solved by reinstalling SD/Dreambooth.

[@anonymousmuskox1893]

I figured it out - make sure the entire folder path that has stable-diffusion-webui has NO spaces in it.

[@yatobatoshu8299]

@@anonymousmuskox1893 you are godsent! Can you change the name without affecting the file structure? I was considering to remove the spaces but I thought it would break my installation

[@johndoefpv]

@@yatobatoshu8299 i just changed it and had no issue whatsoever

[@simonavittoria586]

@@anonymousmuskox1893 THX 🙂. You are right :) I had space in my folder path. I deleted the space and now the installation works fine. 👍👍👍

[@LoneRanger.801]

Hey, what’s your tech setup? I’ve been struggling getting WSL2 to talk to docker (which must talk to CUDA).

[@sebastiankamph]

I run Windows native. My Linux days are far behind me. 5800x and a rtx 3080.

[@kamus2056]

thank you very much..!! 😀

[@sebastiankamph]

You're very welcome buddy!

[@Jdeadevil]

Is there anything that tells you how long it will take if you use your CPU? Would be interested in trying this, as I've got a Ryzen 9 3900XT

[@TonyStark-ik9em]

The ETA will be shown in the GUI and also in the shell.

[@DanielLudes]

Thank you so much, I've learned so much over the last few month about creative ai and you are always one of the first sources of knowledge as soon as a new thing comes out! I'm thinking of getting a new GPU, what are you using, can you recommend one?

[@sebastiankamph]

Awesome, thank you so much for the kinds words! 💎 I'm using the rtx 3080 and it's fairly fast, however I would have wanted a little more vram, preferrably over 12gb. But unless you care about locally training Dreambooth models, something similar to 3080 is great. I get 10+ it/s. More with xformers, which is decently fast. The new 4070 and 4080 would probably be twice as fast. Stick to Nvidia though. Easier with Stable diffusion.

[@DanielLudes]

@@sebastiankamph Great, thanks for the reco! 4080 with 20 it/s would be .... fast, jesus!

[@birknygaard2944]

This seems to be pretty straight forward, and it does install, but afterwards there is no tab for dream booth. Any idea why?

[@tivadarczondi2514]

same to me

[@birknygaard2944]

@@tivadarczondi2514 fixed it. Reinstalled A1111 and followed this, voila

[@birknygaard2944]

buuut... then when trying to run, it fails. get this: Error completing request Arguments: ('withMe', 'C:\DELEMAPPE\Trenemappe\Darkr\512x512\jpg', '', 'darkr', '', '', '', 1.0, 7.5, 40.0, 0, 512, False, True, 1, 1, 1, 1200, 1, True, 5e-06, False, 'constant', 0, False, 0.9, 0.999, 0.01, 1e-08, 1, 500, 0, 'fp16', True, '', False, True, True) {} Traceback (most recent call last): File "C:\Users\Darkr\stable-diffusion-webui\modules\ui.py", line 185, in f res = list(func(args, **kwargs)) File "C:\Users\Darkr\stable-diffusion-webui\webui.py", line 54, in f res = func(args, **kwargs) File "C:\Users\Darkr\stable-diffusion-webui\extensions\sd_dreambooth_extension\dreambooth\dreambooth.py", line 265, in start_training trained_steps = main(config) File "C:\Users\Darkr\stable-diffusion-webui\extensions\sd_dreambooth_extension\dreambooth\train_dreambooth.py", line 872, in main save_model = not global_step % args.save_embedding_every ZeroDivisionError: integer division or modulo by zero

[@bobdrozd6680]

@@birknygaard2944 Sorry, what is A1111? And how to reinstall it?

[@sebastiankamph]

kzbin.info/www/bejne/ennEfWhshZuZa68

[@Mikes-Code]

If I have 8GB VRAM, will it not work or just be very slow? Should I in that case switch to CPU instead?

[@sebastiankamph]

I'd recommend google colab instead. Check TheLastBen

[@Jojo2]

Things I need answers for: How do you find what an existing class/prompt is called? If you name it the same thing, will it merge your results with the previous, or overwrite them?

[@sebastiankamph]

1. You can't if you forget 2. Merge

[@walidflux]

colab still tops

[@brexitgreens]

Why not make a HuggingFace Space?

[@jourdanbiziou853]

Very cool :) do you know if you can Train it on 2 GPU ? I have a Dual 2080 ,,, just wondering, thanks for the info : )

[@sebastiankamph]

I know it wasn't possible a while back, but not sure if they've managed to sort that yet. I'd guess no, but it moves so fast now so who knows 😊

[@jourdanbiziou853]

@@sebastiankamph Hey Sabastian :) thanks for responding, yeah,, im not sure as well,, If I learn of anything I will be sure to share that with you :) thanks again for all the great information,

[@Arcticwhir]

So do you recommended not using a Class prompt?

[@digitalflick]

awesome, do you mind sharing the prompts you used to get your images?

[@sebastiankamph]

Join us in Discord and ask there and we'll sort it out.

[@brexitgreens]

Don't join Discord. It's a ruse. They will kidnap you and use you as AI training material.

[@sebastiankamph]

@@brexitgreens How did you know!?

[@brexitgreens]

@@sebastiankamph 😏😁♥️

[@massive-duck]

hey, i was wondering what your pc specs are

[@sebastiankamph]

Hey! I use a Ryzen 5900x and a RTX 3080.

[@cquencegroup]

Hello! Thanks for the tutorial. I have followed the steps but when I install the Dreambooth extension, it just won't show up even if I reload the stable diffusion web UI. Any fix? :(

[@CoconutPete]

so difficult as dreambooth has changed since a year ago so tough to figure out settings.. will keep trying lol

[@ramirezkine5704]

Hold on, does this mean that you no longer need a gpu for generating images?

[@devnull_]

More like there is now at least an option, if you want run something without a capable GPU.

[@sebastiankamph]

Yeah, what devnull said. The option exists, but having a good gpu is the preferred choice.

[@ramirezkine5704]

@@devnull_ Do you know if I can run SD in a gtx 1050ti? It's a 2gb vram gpu but i couldn't find any information related with text to image generators

[@HertWasHere]

Ive got 6GB and it seems to be doing fine

[@everythingis3d.-lofi-.113]

I have multiple gpu's on one system will I be able to use the combined vram of them for this?

[@Azzalto40]

and how to add dreambooth on Stable Difusion??

[@begumtrump6950]

Can it be used to train faces of 3 people? How should one go about it? Thanks

[@Skullknightjay]

What number for classifications images do I put? If the AI already has dogs, I assume I put the number of pictures it used to train "dogs" but how am I supposed to know how many it used?

[@Skullknightjay]

(I want to add my own dog pictures to the existing dogs in SD)

[@exuberance5757]

In regards to the error due to spaces in your folder path, how do you rename it without breaking your path or reconnect your path? When I rename mine, SD no longer runs

[@DeliciousBrain12]

THIS is what I want to know, too :D

[@romulo4804]

Can my 3070 8gbvram handle it? Even if i will need to waiy for 2 days for it?

[@sebastiankamph]

Sorry, no :( Go check TheLastBen's or Shivam's colab (see links in description). You can run CPU though, and that will take like 2 days. But colab is faster and better.

[@romulo4804]

@@sebastiankamph ok man, thanks.

[@ysy69]

hi, when using the class prompt, do you add such prompt into the instance prompt ? Example: instance prompt: myreflex person class prompt: person Re: regularization images, do all image files need to be named "person(X).png" ?

[@sebastiankamph]

No, the instance prompt shouldn't include the class prompt while training. That's set separate.

[@ysy69]

@@sebastiankamph how about at inference?

[@NapalmCandy]

Please! Add a guide to run it locally on 10Gb VRAM GPU, Automatic1111 or not, but locally. Thanks!

[@motovlogsipanca4832]

by stressting out the gpu a lot , is there anything i can do to not getting fragmentation?

[@xellostube]

For some reason it doesnt save any ckpt any advice?

[@CaseysCountdown]

Helli mister! I have a 4gb vram 3050RTX in a laptop, which gets me to a CUDA error, tries to allocate 58mb more, which i dont have obvs. But i can't find that option in the dreambooth to use CPU istead, any ideas why? Thank you in advance

[@chaitanyamore8786]

Do this work on amd?

[@MitrichDX]

Где взять старую версию? Обновился на какую-то от shivamshirao и это ужас какой-то скорость хуже, вкладок, и пунктов попросту чересчур много.

[@sebastiankamph]

Just delete your Dreambooth folder. Then git clone the github version you want.

[@MitrichDX]

@@sebastiankamph удалял, git pull восстанавливает то же что и было( скиньте кто-нибудь у кого в папке Extension еще лежит старая версия на 2 вкладки, а не на 8 =(

[@realityzoner]

Any idea where i can download class images from ?

[@ac3d657]

generate them.

[@realityzoner]

@@ac3d657 it's okay i found some pregenrated

[@adrianmucci4860]

Can I use 1660ti for this? I don’t care the time. Only need to work. I have a ryzen 3 3100 16GB ram and 1660ti. Thanks for sharing.

[@sebastiankamph]

No, you will have to run on cpu only. Will take something like 24 hours. I would recommend running TheLastBen's colab in video description. Takes 30 minutes to get a .ckpt.

[@ZEK1NRZ]

i think your source model you're still using 1.4, any reason for that?

[@TaoZen888]

is anybody already figure out to use gpu also ?

[@seikkv4371]

U cant made this work on 8gbvram even with --lowvram in orders? And another question, how many tame take to train a model like your example of 12 images? (And with CPU only if somebody knows?)

[@sebastiankamph]

I don't think 8 will cut it. I still use Google Colab most of the time, takes about 30 minutes for that amount of images.

[@drorlazar]

gr8 tutorial! thanks when I follow it, after creating the model and right after pressing "train" I'm getting this error: "CUDA kernel errors might be asynchronously reported at some other API call,so the stacktrace below might be incorrect. For debugging consider passing CUDA_LAUNCH_BLOCKING=1." I have a 3060ti with 12Gb, it should be able to run on 12 images on 500 steps (as a 1st test) no? any ideas?

[@sebastiankamph]

That's an error I haven't seen before. Check the github for issues. Sorry man, it's a little buggy. Thanks for your support though ☺

[@lalamax3d]

thanks, few questions. 1- if i have images of 20 ppl in 20 directories. how can i train them with 20 prompt words in one go, i.e loading different model for each one is hard. thanks for your time.

[@sebastiankamph]

I don't think that's possible yet. But you can create 20 models with 20 prompt words and then merge them together to 1 model. I can't guarantee the results will be great with so many, but I've gotten it to work with fewer.

[@simonduque9943]

Hey folks, I finally was able to install the Dreambooth extension and train a model with my pictures. So, how do I actually use it to general new AI art with my face?

[@simonduque9943]

Don't I need a .CKPT file to add to the modules folder? I only see .BIN and .JSON files in the output folder

[@simonduque9943]

Also I'm getting only 100/1200 steps, then it says "training completed" which is weird, not sure if it's actually an error. I have an RTX 3090 so there shouldn't be any hardware problems

[@sebastiankamph]

I've seen a bug where training can complete or bug out if the preview setting is 0 or lower than the training steps. Could be that.

[@lospuntosstudios5149]

Why is Midjourney V4 so perfect? Will Stable Diffusion be that good the next weeks?

[@stefanb7015]

MJ got worse - it just generates the same person over and over again, 90% of the images look identical now unless you go with high --chaos.

[@lospuntosstudios5149]

@@stefanb7015 kzbin.info/www/bejne/bHekmn1-oZmLrck

[@ShamusMac]

@@stefanb7015 Add the flag `--v 4` to use MidJourney 4

[@beecee793]

I laugh when I see people say this. MJ is only as good as it is because of SD lol

[@planetdamageofficial]

Hey everyone - I'm pretty much stuck with a "tuple index out of range" error - the model is created, but apparently I just can't get the training process to start. Can anyone please help me out with this?

[@rajeshkannanmj]

It is weird that I have to use VPN to load things from "Load from". Does it block countries? or is it to me only? I am from India.

[@jamesmaynardjr]

Can anyone help me? I installed Dreambooth this same way but some things just say error and won’t work

[@aiarchitecture6427]

I am confused, title says " CPU only option " and after we recommend 12Go Vram minimum ?

[@blimeyitsRichard]

there's a 'cpu only' checkbox in the trraining parameters section

[@_DARTH2602]

dreambooth no longer installs and starts correctly in SD web ui (google colab). impossible to use it

[@sebastiankamph]

Sorry to hear that! I'd recommend trying one of the stable Google colabs, like TheLastBen's.

[@FLEXTORGAMINGERA]

I am still concern if my pc is ready to train . My specs are ( i5 8400, 32gb ram and i5 8400 8th gen cpu )

[@sebastiankamph]

Unless you have 10-12+ gb vram I would recommend to create a .ckpt from a Colab, like TheLastBen's (see video description). That's a faster option than running cpu only in auto1111. But at least you have the option now.

[@SjonSjine]

RTX 3080 with (only) 10Gb vram, did not work, errors - now trying the CPU only but also problems - What would be the settings for CPU only? Please advice!

[@sebastiankamph]

Honestly, since you're on a highspec computer, I'd just go to TheLastBen's colab, run a training there for 20 minutes and then import the .ckpt. Cpu is super slow.

[@newnigerianmovies]

can dreambooth automatic 1111 be hosted on an azure or aws gpu server?. if yes how

[@YT_Whitey]

Is there anywhere to get technical help on this? I follow all the steps but the training at the end doesn't happen. I tried the SD discord but no one could assist.

[@sebastiankamph]

Check my discord in channel description.

[@YT_Whitey]

@@sebastiankamph Thanks, will do!

[@adamharp2797]

I have a laptop what video should I use an could you help me out thx

[@sebastiankamph]

Install Stable diffusion automatic1111 (Ultimate guide). If you want Dreambooth, use TheLastBen's colab for ease of use. Shivam's for a little better result.

[@adamharp2797]

@@sebastiankamph when I did it it was talking about Vae