Realization LoRA.
It is illustrated when operated in "minus".
リアル化LoRAです。
「マイナス」で動作させるとイラスト化します。
With your support, I can continue to develop.
You can also participate in the testing of the model under test.
皆さんの応援があれば、私は開発を続けることができます。
また、テスト中のモデルのテストに参加することもできます。
BreakDomain (illustration output models )
https://civarchive.com/models/50520/breakdomain
BreakDomainRealistic (illustration output models that are more realistic)
https://civarchive.com/models/54917/breakdomainrealistic
BreakRealize LoRA (LoRA that adjusts to a realistic representation)
https://civarchive.com/models/61810/breakrealize-lora
BD_Marumi LoRA (LoRA to create an animated expression)
https://civarchive.com/models/70877/bdmarumi-lora
Description
First Release.
FAQ
Comments (8)
It is more habitual than you can imagine. It's not just a level of realism or illustration.
It's a short prompt with only 1girl, and it's for "people who want to splash Tabasco or honey on it to change the flavor." I wouldn't stop.
It works for about 60 tokens, not too long or too short." This is for "people who want to add pepper and herbs to change the flavor. I also add red ginger.
想像以上に癖がある。リアルとかイラストとかいうレベルじゃない。
1girlしかいない短いプロンプトで、"タバスコやハチミツをかけて味を変えたい人 "のためのものです。私なら止めません。
長すぎず短すぎず、60トークンくらいで有効です。" これは「胡椒やハーブを加えて味を変えたい人」向けです。私は紅しょうがを入れたりもします。
Oh wow, thank you! We were just chatting about BREAK on your checkpoint BreakDomainRealistic. Can you please describe what this LoRA does and how it is changes the BREAK command? It's not clear from your description or examples what is going on exactly. Google Translate only goes so far :)
the "BREAK" is for when someone is using Regional Prompting extension in A1111.
@KinkMaster No, it is a default feature in A1111. Read the below comment thread for more details, it contains links with A1111's vague explanation and a more in depth review that you can translate with Google Translate (install in browser, right click page > translate).
https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#break-keyword
Still not exactly clear how this Lora is affecting it though.
Hello.
I am using ChatGPT to translate this message as I am Japanese and not very good at using English. I apologize if there are any mistakes.
BREAK
BREAK seems to prevent words before the prompt from affecting the words after BREAK in the text by forcibly filling in 75 tokens. For example, if you input "green eyes, prompt..." the generated image will have green mixed in not only in the eyes but in the entire image. The prompt within 75 tokens seems to affect the following words. However, if you input "green eyes, BREAK, prompt...", BREAK separates "green eyes" and "prompt..." within the same 75 tokens, so the prompt can be prevented from turning green.
For more details, the author has published an article in Japanese. The translation may make it easier to understand.
https://br-d.fanbox.cc/posts/5731955
LORA
I am not sure about the exact behavior of LORA, but recently in Japan, there has been a technique where setting the strength of a flat color LORA to -1, like in anime style, can make it more realistic or increase the detailing. Here is an example:
https://twitter.com/siroroai/status/1647466318264934402
It is likely an application of this technique.
from what I can tell, this LoRA skews the result towards the beggining of the prompt or towards the end of the prompt, depending on the weight - that's pretty darn cool!
@charlesp thanks! now i understand what this is for. also, kudos to the authors
@charlesp So then in theory, you could write one massive prompt, separated by a bunch of BREAKs, generate a grid like the creator has done here with weight from, for example, -1 to 1, and generate a series of frames for an animated GIF using a single prompt and a single tap of Generate, without relying on any extensions beyond what's build-in to webui? And not only that, the frames appear to be a much smoother transition than the stuff I've seen coming out of most of the video extensions. You're right; that's pretty cool indeed.