doge

A multi-concept LORA model trained on ~60 images of doge characters and shibas in different styles, which makes it compatible with almost any other 1.5 model.

This LORA includes the following characters: doge, cheems, swoledoge, baby-doge + bonus characters and generally improves the ability to generate shiba inu dogs.

Optimal CFG Scale 3-6
Optimal Lora ranges: 0.55 - 1.1

To use this model In a1111 webui, download it, rename it to doge.safetensors place it in models/lora and include <lora:doge:1> anywhere in the prompt, where 1 can be any value in lora ranges mentioned above. Click on the images to see the prompts used, if it doesn't work, turn off ublock or refresh the page.

The trigger phrases are: doge (for the main doge character), shiba inu (for more general shibas), buff Swoledoge (for bodybuilder doge), baby-doge (for a cute mini-doge puppy), Cheems sitting (for the Cheems character).
For Cheems you might need to add, "puppy p3rr0" to the negative prompt.

Bonus characters: this lora can also generate two other dogelore characters: Perro and Walter. I found a consistent recipe for Perro for realistic models, which requires a warm-up stage to give the basic shapes and colors. The prompt requires a trigger phrase [(white) puppy:cute ((p3rr0)):5] sitting where 5 is ~1/4 of all generation steps; CFG scale 3-4; lora power at 0.7-0.8, and "(doge)" and "(shiba inu)" in negative prompt.

For Walter, it is a bit more simple: white (bullterrier) ((walt3r))

Why did I use leetspeak in trigger words during training? Because both "perro" and "walter" are very popular words/tokens that do not match the expected concept.

If you like this lora, feel free to follow me <3
Message to community: STOP making 144mb LORAs for a single character.

License: use as you wish, excluding usage of this free model for cryptocurrency promotions, it is prohibited under default license and requires a commercial license, which you can obtain by contacting me.