The main goal of the LoRa is to introduce a wide variety of furry species so you can actually prompt for furries while drawing on Anima's knowledge.

My typical gen settings: 30 steps, cfg 5, er_sde sampler, simple scheduler. Trigger is "furry". Include anywhere in your prompt.

Typical anima resolutions work, you can push them higher with the official highres lora https://civarchive.com/models/2540444/anima-highresaesthetic-boost

Please also read the Anima documentation. lol.

other notes on the training/usage

• trained on Anima Preview v3

• Dataset is 55k images of species with above 300 tags on e621. A list of trained species can be found in the huggingface repo https://huggingface.co/garaggge/Anima-Preview-v3-General-Species-LoRA/tree/main

• Trigger is "furry". Include anywhere in your prompt.

• Prompting species prior to artist tag seems to help with species adherence, doesn't seem strictly necessary though.

• Training was conducted with e621 tags, so a lot of e6 concepts/tags are possible now. Don't expect miracles with niche fetishes.

• Tags were deimplicated, so you don't have to prompt "arthropod, insect, lepidopteran, moth" to get a moth for example, you can prompt just "moth".

• Don't use underscores in your prompts (lol)

• Prompting "1girl", "1boy", etc. may bias towards a more Anime/humanoid look, but it isn't always overwhelming in my testing

• Only species, general, and character tags were included. Tags were shuffled in training.

• Artist tags were not included

• Anima Artist tags still work (@artist)

• Natural language works to varying degrees