Anima image model tips: Stabilize artist tag mixing with @, weights, and shift

The Anima AI image model, designed for flexibility, often produces wildly inconsistent results when mixing multiple artist tags—varying style, realism, or eye appearance from seed to seed. A detailed community post shares a systematic approach to stabilizing these mixes. Key techniques include: prepending the @ symbol to artist tags (e.g., @anime coloring) for more consistent style application, increasing the weight of the entire artist block (start with (:2.0)), and then reducing individual artist weights for fine-tuning. The shift parameter should be raised for more tags—values around 10-24 work without breaking the image. Proper spacing is critical; missing a space after a comma before a character tag can cause the model to ignore it entirely.

Additional advice focuses on tag selection: avoid 'masterpiece' and high Pony scores as they overpower styles, and be careful with positive and negative prompts that carry inherent style bias (e.g., 'dot nose' implies flat color, 'nose' and 'lips' shift toward realism). Users who prefer not to manually craft tags can rely on LoRAs, but many existing LoRAs are poorly tagged or overbaked—detectable by characters always looking away. The community emphasizes that Anima is not aesthetically fine-tuned; it's meant for easy training, so prompt engineering is essential for consistent output.