How To Create a LoRA of Yourself with AI Using Stable Diffusion

[@grobknoblin5402]

This is hands down teh best lora video ive seen tried a load of others and never got it to work. You are amazing! I cant wait to see more of your videos. You explain it all clearly each step of the way thank you! definately earned a subscription!

[@BlaizeTech]

Glad it was helpful!!!

[@Ialreadyhaveaname2]

Sir, you are a lifesaver, thanks for this.

[@BlaizeTech]

I'm glad silly images can be a life saver. :D

[@inctube2320]

Thank you for this walkthrough! Can't wait to start training my own lora

[@BlaizeTech]

Glad you found it helpful

[@cyberspider78910]

This is an excellent tutorial to begin on LORA trailing...

[@BlaizeTech]

Kind of a fun and easy intro for it.

[@GeorgeSmith-l8b]

Thank you - that was a great walk through.

[@servermatter4465]

How would you create a video representation of this using a script where it mouths word of the script you provide

[@BlaizeTech]

I'm not sure how to do that like that. :/

[@Aversy]

great tutorial, can you tell me more about the VM? i also has a terrible laptop so im considering using a vm.. how expensive is it?

[@BlaizeTech]

I use spot instances on azure VMs, which means it costs about $0.10-0.35 for the VM. I also shut it down when I'm not using it. In a given month, that costs about $10-20. But that all depends on what you want to do.

[@cyberspider78910]

I had used VMWare virtual machine which you can install on standard window. This basically creates PC with in PC. The thing I don't know is whether it support CUDA simulation or not?

[@BlaizeTech]

@@cyberspider78910 VMWare is tricky. It can do some things on the GPU using passthrough or vGPU, but it's not always supported by the hardware. You'll have to check to make sure your machine works for that situation. I'd run it on the host in any case if you can.

[@vpgms]

when training it does not let me cache latents (0/25) and then traces back and ends the training soon after, leaving me with a toml file and not the safetensors. I dont know how to fix this issue

[@BlaizeTech]

Check your permissions and drive space on your computer where you're training the images. It sounds like it's unable to write files it needs.

[@vpgms]

thank you I will check

[@akarimatsumura6601]

thanks you so much for the video, i created my own lora with base model SD 1.5 successfully, however there are many sampling method, DPM++ 2M, 3M, euler, heun...etc, how do i know which sampler work best with my lora? Do i have to test it with different sampler,sampling steps, CFG Scale and weight to find out? is there any method to find out without testing it one by one? thanks

[@BlaizeTech]

Which one works best is subjective... Finding the optimized settings for a LORA seems to be trial and error. I don't think there is a "best" one. Turning up the sampling steps helps some. The other trick is to to use the keyword in your LoRA with manual weights (word:weight) ie (blaize:1.3) and also amp up your LoRA or something like that.

[@marcelbeijer1]

Nice tutorial. One question. In the Training tab I see no Parameters section. I use v 24 1.4. In that version Under the Folders section follows Dataset Preparation. In your tutorial there is a Parametes section. How can I fix that?

[@DEVANK-gb7ed]

Hi, are you able to train your LoRA using this tutorial ?

[@marcelbeijer1]

@@DEVANK-gb7ed Not yet, because I don't see the Parameters section appear. But on to that point, the tutorial was good

[@BlaizeTech]

@@marcelbeijer1 they changed the UI. Look under utilities for the captioning tools, then the rest of it should be the same.

[@Tallsquall]

I missed it if it was specifically called out - do all the pictures need to have the same aspect ratio?

[@BlaizeTech]

No, they don't. variations are probably better than what I have in the video.

[@Frontlainer]

What GPU do you have?😊

[@BlaizeTech]

3060 TI. I did those on an Azure VM, though.

[@Misslee06]

thank you

[@BlaizeTech]

You're welcome

[@mosh4741]

"that took about 20 minutes" his clock moved 1 minute lol

[@BlaizeTech]

The magic of editing. :D

[@chriscannon4195]

im not getting anywhere with this. training lasts like 3 seconds and get no lora

[@BlaizeTech]

Do you get an error in the console?

[@danhitchner]

I'm geting an error. It looks like it is unable to find the SDXL preset. The message is, "FileNotFoundError: [WinError 3] The system cannot find the path specified: '/kohya_ss/output/SDXL1.0_Essenz-series-by-AI_Characters_Concept_Morphing-v1.0'"

[@BlaizeTech]

You might wan to try and reinstall kohya again. Sometimes, it messes itself up on an update.

[@SirOlf-i7n]

The title is kinda misleading since you use Kohya and not stable diffusion

[@BlaizeTech]

Maybe. Kohya provides a suite of tools for training, fine-tuning, and optimizing Stable Diffusion models, so I don't think it's misleading.

[@SirOlf-i7n]

@BlaizeTech the title States "creating Lora using stable diffusion" but you are not using stable diffusion at all to make the Lora.

[@willianderuberkater3526]

You should give kwebbelkop ai a try 😂

[@BlaizeTech]

😂😂

[@Misslee06]

thank you