Tansan

(Japanese, noun, 'tahn-sahn')

1. carbonic acid​; 2. carbonated water; carbonated drink​ (abbreviation).

Alright, I needed something light and refreshing after my last LoRA. 😂🙏

Tansan is a hand-drawn Anime portrait composition LoRA. I know, I know, there's like 50 million anime LoRAs available already. I'd like to think this one is a little special though.

The style is primarily focused on portrait composition - the composition and the art were linked in the training process to create a LoRA that has a very specific use, but does it well. In particular, this LoRA excels at specific-focus, depth scaling, dynamic poses, floating objects, and flowing elements.

This LoRA craves depth scaling. I have actually had it refuse to work in one or two instances where it couldn't create depth - like I said, the art and the composition are tied together. But it works so well, it's hard not to give it what it wants, creating a transitional blurring of foreground and background objects to create focus in the mid-ground, where the action is.

The floating objects and flowing elements feed into the depth scaling. Think scenes with cascading energy, flowing liquid, flying bullets, or mundane objects suspended for surrealist effect. It's highly stylised and tends to capture a beautiful flowy mishmash of action and motion in still portrats with minimal prompting - no trigger word required. 🙌

This LoRA was not made to be NSFW ready. It does return elements of nudity sometimes (despite not being trained 🤷‍♂️), but you can always plug-in your own fav NSFW LoRA in to get some A-Grade smut happening. 👌

So, here it is. A cutesy fun LoRA in an overused style that hopefully brings something a little different to the table. This was my first attempt at a style LoRA, so has been an amazing learning process.

Fun facts - Tansan was originally a brand of Japanese carbonated water, and is used as slang for bottlecaps in the Phillipines. 🤔☝️😅

Next LoRA will be a little bit more offbeat I think... 😈

ZiT Version Notes

This version is good for anime portrait composition, but struggles with complex dynamic shots or multiple characters. Best use is single focus character images. It has great use of depth scaling and loves putting depth blur on foreground and background with a strong focus on the mid-ground. It is very strong and operates best at about strength 0.8, though you can drop to 0.6 or lower if you're struggling to get a consistent shot.

Qwen Image Version Notes

This version operates quite well at 0.8 strength, you can go lower, however you lose a lot of the hand-drawn look and detail if you do. It has a high level of fluidity with flowing elements and dynamic poses, however anatomy is more of a fluid concept to this version as a result. It sometimes gives weird anatomical anomalies, especially hands and feet which can easily get swept up in its artistic flair. Lowering LoRA strength helps offset this issue. The other option feels a bit dated, but the old 'best hands, five fingers, good anatomy' prompting can assist also.

Enjoy! 😁👌