my pc doesn't have enough vram, so i paid for google colab compute GPUs. i guess if i pay for runpod
my pc doesn't have enough vram, so i paid for google colab compute GPUs. i guess if i pay for runpod then it'll work?
Fyes runpod works
Fi have necessary tutorials
FGitHub
Best Stable Diffusion and AI Tutorials, Guides, News, Tips and Tricks - GitHub - FurkanGozukara/Stable-Diffusion: Best Stable Diffusion and AI Tutorials, Guides, News, Tips and Tricks
Fsearch runpod word
Fi am going to sleep hopefully see you tomorrow
Kthank you! is there a way to use google colab to do it? just because i already have some compute on google colab and don't want to waste them
Tme: i don't really understand truncation error with regards to image diffusion samplers. can you lay it out in eli5 terms for me?
ChatGPT said:
Sure! Let's say you're baking cookies and the recipe calls for 1/3 cup of sugar. But your measuring cups only have 1/4 cup and 1/2 cup.
If you use the 1/4 cup, you won't be adding enough sugar. If you use the 1/2 cup, it will be too much. So, you decide to fill the 1/4 cup and add a little bit more, hoping it's about 1/3 cup.
The difference between the amount of sugar you actually added and the exact 1/3 cup the recipe called for is like the truncation error. It's a small error that comes from having to stop (or "truncate") your measuring because you don't have the exact tool (in this case, the right measuring cup) to get the precise amount.
In the case of the image diffusion, the "sampling steps" are like the measuring cups. If you could do an infinite number of steps, you could get the exact, perfect result. But in reality, you have to stop at some point (just like you can't keep adding sugar forever), and this results in a small error. Increasing the number of steps is like adding a bit more sugar to the 1/4 cup to try and get closer to the exact 1/3 cup you need.
ChatGPT said:
Sure! Let's say you're baking cookies and the recipe calls for 1/3 cup of sugar. But your measuring cups only have 1/4 cup and 1/2 cup.
If you use the 1/4 cup, you won't be adding enough sugar. If you use the 1/2 cup, it will be too much. So, you decide to fill the 1/4 cup and add a little bit more, hoping it's about 1/3 cup.
The difference between the amount of sugar you actually added and the exact 1/3 cup the recipe called for is like the truncation error. It's a small error that comes from having to stop (or "truncate") your measuring because you don't have the exact tool (in this case, the right measuring cup) to get the precise amount.
In the case of the image diffusion, the "sampling steps" are like the measuring cups. If you could do an infinite number of steps, you could get the exact, perfect result. But in reality, you have to stop at some point (just like you can't keep adding sugar forever), and this results in a small error. Increasing the number of steps is like adding a bit more sugar to the 1/4 cup to try and get closer to the exact 1/3 cup you need.
Thow correct did chatGPT get this i wonder?
生hey guys i try this vedio but the cmd show this is means i dont process in GPU? my gpu is RX6600
生YouTubeSECourses
Databricks’ #Dolly v2 is a free, open source, commercially useable ChatGPT-style #AI model. Dolly 2.0 could spark a new wave of fully open source LLMs similar to #ChatGPT . Open source community working hardest to bring up a model that can compete with GPT4. Our discord: https://bit.ly/SECoursesDiscord
If I have been of assistance to you and yo...
If I have been of assistance to you and yo...
生
Ti'm a nooby, but from what others have told me, AMD doesn't work well with machine learning stuff like stable diffusion (yet)
生oh that too bad QQ
Ti'm not familiar with the specific error message your seeing, so maybe someone can give better help. i'm not saying AMD can't run stable diffusion
Tjust that it isn't as stable as an nvidia card or as easy
生alright i will try this again with my own computer
生thx
Tgood luck
AWas wondering if there is any way to instruct SD (A1111) within prompt to only apply lora to specific subjects? Say I have a lora that generates specific person, if I put few subjects it will usually apply lora to them to all, is there a way to limit lora's influence to only one specific subject without having to use regional prompt or other similar extensions?
Uyes there is, you need to learn to use latent couple and masking
Ualso if people werent aware https://discord.com/invite/stablediffusion
Utheyre letting you test and vote on SDXL (reportedly 50% done)
Uive 3 prompts currently in the current round of voting which is neat
UI see now you are asking without using those extensions. your best bet becomes inpainting i guess.
Ui really recommend latent couple tho since it lets you create masks and define each one in a single prompt, ill give example:
Uusing control net plus latent couple i could define all 6 characters. i can bring them out a lot more with more control nets and defined details
Uthetres a lot of room for flexibility with the modified softedge_hedsafe control net im using
Uthe bottom right image is the latent couple mask, the colors are arranged in order that they'll fill out in the form underneath, white being the general prompt for scene with left to right in color being each character. you can do it with loras. there was a bug for a bit using loras in prompts, but i think tehy fixed it? so if you were to make an openpose control net and latent couple each pose similar to how my masking was done, your prompt would read:
generral scene of scene-like things and such BREAK
AND <lora A:0.6> BREAK
AND <lora B:0.6> BREAK
AND <lora C:0.6> BREAK
AND <lora D:0.6>
generral scene of scene-like things and such BREAK
AND <lora A:0.6> BREAK
AND <lora B:0.6> BREAK
AND <lora C:0.6> BREAK
AND <lora D:0.6>
Uyou have to use those steps tho if you want to do what you're talkingabout in one generation
UI am not using loras for my image, but this is the only way i know how. i know you wanted a different answer but hope this might help
Ubetter render and more info
Ffor google colab it is not working as good as on your pc
FChatGPT has no idead about stable diffusion yet. but i am planning an in depth tutorial for this
Fhi. probably they dont support amd right away. there could be way to make it work but i havent tested
Fyou are using latent couple extension? this is something i plan to cover
UYes. I love the feature of defining a mask as it allows. I've not tried regional prompter or whatever it's called, I looked at it and it seemed to be a step backwards in function
UTotal power over generations between latent couple and control net, tho to fully utilize latent couple, requires some skill with Photoshop or similar for precision. The control net segment preprocessor might have some potential too. I've not tried it just considered it.
AI aim finding that I get better training results using
linear as the learning rate scheduler than using constant_with_warmupUPersonally cosine and polynomial have given me good results. Seems to be a matter of just finding settings. There's a write up in discussions for dreambooth on GitHub that has a very thorough fleshing out of how to do your settings according to a formula.
https://github.com/d8ahazard/sd_dreambooth_extension/discussions/547
https://github.com/d8ahazard/sd_dreambooth_extension/discussions/547
GitHub
NEW UPDATE Updated formulas for calculating frame resolutions. Also updates with permissions were carried out in the table with mathematics. The custom lr_scheduler has been moved to a separate doc...
AThanks, any chance you could share your settings with me?
Fnewest xformers is compatiable with torch 2.0.1
FYouTubeSECourses
RunPod: https://bit.ly/RunPodIO. Discord : https://bit.ly/SECoursesDiscord. Web UI DreamBooth got epic update and we tested all new features to find the best settings (10). 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...
https://www.patreon.com/SECoursesPlaylist of...
ABuy these are all constant_with_warmup aren't they? and polynomial and linear both work better.
U@ashleyk I would not presume to know better than the good doctor.
I am mobile at this time and cannot access my dreambooth settings. The advantage I like of any scheduler on a curve is that it greatly helps with avoiding over fitting. Lots of ways to preserve the model you train on.
I am mobile at this time and cannot access my dreambooth settings. The advantage I like of any scheduler on a curve is that it greatly helps with avoiding over fitting. Lots of ways to preserve the model you train on.
UBut I cannot argue with the doctor's results.
UIt's 6am here and bed time, but as I need to get back into training this weekend, ty for the link. I will definitely watch. I've been bookmarking several of your courses for references.
AThis video is awesome, you definitely need to watch it if you have not already done so
UI plan to. His courses are fantastic. This weekend I have a lot of work I need to get done, so I'll be training to his spec and looking forward to the results
UI am really putting my 3090 TI through the ringer! Wish I had a 4090 but I can't be unhappy with what I do have.
AYeah you are fortunate to have your own GPU, I don't have a GPU so use the cloud for training
