Has anyone gotten training with #LoRa to to work in #Dreambooth? I just get noise for the preview images. I'm guessing I need to turn the learning rate way down?

Also, is the LoRa that gets saved actually useful with stable-diffusion-webui, or is it only useful for speed and VRAM savings, and one still needs to use something like https://github.com/bmaltais/kohya_ss/blob/master/networks/extract_lora_from_models.py to extract a useful LoRa?

#StableDiffusion

@stablehorde
Do you know anything about #DreamBooth free access?

#stablediffusion #dreambooth #deforum #ai #AIGenerated

Another collaboration with visual artist Lili Fiallo. Using #dreambooth we fine tuned #stablediffusion to incorporate Lili's toy art style. incredibly simple using diffusers and gradio libraries.

https://instagram.com/fiallolili/
https://3e2bb3f1180757bdee.gradio.live

Ever wondered what it looks like to create high quality printable works of art with Stable Diffusion?

I just published my latest blog post about how I used Stable Diffusion and Dreambooth to create a "painted" portrait of my dog 🧀 Queso

https://www.shruggingface.com/blog/how-i-used-stable-diffusion-and-dreambooth-to-create-a-painted-portrait-of-my-dog

#art #aiart #stablediffusion #artist #dreambooth

As far as I can tell, basically every popular tutorial about how to fine-tune #StableDiffusion models using #DreamBooth is chock full of bullshit. If any of the people posting these ever bothered to read the paper, they didn't actually understand much of it.

In particular, everyone seems to misunderstand what "classification"/"regularization" images are for. The first name is confusing; they should just call them regularization images. They're literally just additional images that are used for training so that the model doesn't drift. They tell you to use the "class" of whatever you're trying to teach it (dog, woman, etc), but that's because when you're, say, trying to teach it what your dog looks like, there's a tendency for it to make all dogs look like your dog. So the training process shows it a picture of Fido and says "this is Fido" and then it shows it five pictures of what the model itself already knew are dogs and says "this is a dog." It's that simple.

One bullshit explanation I've seen is that the images keep the fine-tuning from affecting *other* classes. Another is that it somehow trains on the "differences" between your training images and the classification images. But as far as I can tell, the only reason the classification images are treated differently is so that DreamBooth can automatically generate them for you if you don't make them yourself.

Does anyone know how to get #dreambooth working with #xformers in #stablediffusion?

I am in dependency hell, Automatic1111 and Dreambooth keep fighting over xformer version requirements.

Making faces: How to train an AI on your face to create silly portraits - Enlarge / Ever want to be a superhero? We'll show you how. (credit: Sha... - https://arstechnica.com/?p=1924616 #stablediffusion #dreambooth #features #photos #tech #ai

Ars Technica: Making faces: How to train an AI on your face to create silly portraits https://arstechnica.com/?p=1924616 #Tech #arstechnica #IT #Technology #StableDiffusion #Dreambooth #Features #photos #Tech #AI

#aiart #abstract #art #dreambooth

my https://promptdle.com score today is:
🔴🟡🟢
🟢
🔴🟢
🟡🔴🟡🟢
🟢

#promptdle #aiart #stablediffusion #stablediffusionart #midjourney #dreambooth

I love to cooking at home,fresh,organic,fairtrade from local farmers and around the united state of America egg white wrap on a sprouted wheat tortillas,half a grapefruit and a green drink #boyee #pocketorsomejunk #dreambooth #cobbdub #bloweherebruv #squeezecandy #honkywalk #swettybig

Sure, I once supported them because the goal of preventing future trainingdata from being "contaminated" by "bad quality" AI-generated images sounded kinda reasonable to me however thanks to #Dreambooth, etc. making training #StableDiffusion on a few to a few hundred images (meaning a number of images much more easily sortable by humans) possible conventional mass-trainings are now only relevant for proprietary centralized models like DALLE and Midjourney.

After spending the last few weeks exploring how #finetuning #stablediffuson works in #dreambooth at https://replicate.com, I am excited to experiment with fine-tuning an LLM using #flan T5.

I found this walkthrough useful to get started: https://www.philschmid.de/fine-tune-flan-t5

Пытаюсь понять, пойдёт ли мне борода...
Что думаете?

#StableDiffusion #DreamBooth #AIArt #ArtOfArtificial

And a happy #caturday to you all.

https://www.fiverr.com/share/y8d9aG
https://www.fiverr.com/share/wqN1lq

#stablediffusion #stablediffusionart #aiart
#dreambooth
©2023 Brian Ragle

Released datasets of regularization & class instance images for prior preservation loss training: Landscape, space, creature, horse, architecture, and more!

🖼️ https://huggingface.co/3ee

#AIart #generativeart #dreambooth #StableDiffusion #AIArtworks

Занимаюсь созданием искусственного интеллекта....
#artofartificial #neuroart #mastoart #stablediffusion #dreambooth

Обучил Stable Diffusion 1.5 на коте Maxwell из видео "Half Life 2, но у меня есть кот"

https://huggingface.co/kabachuha/maxwell-the-cat-diffusion?text=maxwell+the+cat

#generativeart #stablediffusion #maxwellthecat #maxwell #aiart #meme #halflife2 #garrysmod #dreambooth

I just looked up how much VRAM current gen mid-tier-GPUs (RTX 3060 cards) have (even if I currently can't afford them) and damn, they already have 12GB of VRAM.

I guess we can expect many more forks of #StableDiffusion f.e. finetuned with #Dreambooth, etc. in the near future (at least if Stable Diffusion isn't stopped by #GettyImages, etc).

The open-source AI-revolution has just begun.