/

/

anima

anima

AD

O tym modelu

ANIMA · Modele bazowe

(Updated! Supports Anima LoRAs) This is the v1 base version of Anima, a new model by CircleStone Labs and Comfy Org. The model is primarily an anime/illustration style model, similar to Illustrious & Pony style models from the SDXL era (it is not compatible with those LoRAs). Anima supports both (Gel)booru tags and natural language prompting, its typically recommended to use a mix of both (in any order), but either/or works as well. Prompt weights are supported, but they require much heavier weights than SDXL - for example, (soup:2). Human score style quality tags are supported ("masterpiece", "best quality", etc) along with PonyV7 style ("score_9", "score_8", etc). When using tags, spaces are preferred instead of underscores - with the exception of Pony score tags. When using artist tags, prefix them with '@' such as "@big chungus". Read more at: https://huggingface.co/circlestone-labs/Anima Workflow specific details: This workflow includes a built in upscaler, which can be activated with `/scale:2` (would do a 2x upscale for example). You can control the denoise value for this with `/strength:0.2` (the default). It can also use the CircleStone Labs official Turbo LoRA when you set `/guidance:1` - by default though it uses a guidance of 4.3 Steps by default are 8 when using Turbo, or 16 without Turbo. Their official (non-turbo) step recommendation is 30, but you can typically get away with lower. When using Turbo, expect text to not render as well - can sometimes be worked a