🌟 About This Model: The Perfect Middle Ground

Some Anima models out there suffer from a bad case of extremes. They are either over-baked with harsh, jarring contrast, or completely drained of life with dull, washed-out colors. I put this merge together because I wanted that elusive sweet spot right in the middle—vibrant, clean anime tones combined with a balanced, natural dynamic range.

If you are looking for crisp, sharp, and highly detailed results without the blinding contrast or the faded, muddy fog, you are in the right place (or I hope so lol).

⚙️ Recommended Generation Settings

To get the exact balanced look this merge was designed for, use these parameters as your baseline.

| Setting | Recommended Value | Implementation Notes |
| :--- | :--- | :--- |
| CFG Scale | 3.0 – 3.5 | The absolute sweet spot. Keeps prompt adherence sharp without deep-frying the contrast. Drop to 2.5–3.0 if stacking LoRAs. |
| Steps | 30 – 40 | Ensures high detail and full convergence. 30 for speed, 40 for final renders. |
| Sampler | ER-SDE or DPM++ 2M SDE | Use ER-SDE for flat, precise anime lines. Use DPM++ 2M SDE (GPU Seeded, non-Heun) if you want more creative seed-to-seed variety. |
| Scheduler | Beta / Simple / KL Optimal / SGM Uniform | Baseline standard for clean, artifact-free generations. KL Optimal can be good for a raw detail boost, but it can get wild, so handle with care. |
| VAE | qwen_image_vae | Mandatory. Always use the native Anima VAE to prevent broken colors or dead tensors. |

🚫 Negative Prompt Strategies

Golden Rule for Anima: Less is definitely more. Because this model uses a Qwen LLM text encoder instead of CLIP, massive walls of negative text or parenthesis weighting will completely confuse the model and cause detail collapse. Keep it lowercase, use spaces instead of underscores, and do not use weight brackets.

Choose the tier that best fits your current generation:

🟢 Minimalist (Recommended to start with)

Keeps the AI completely focused and prevents style/detail stripping. Use this for the most natural, unhindered generations.

worst quality, low quality

🟡 Standard (Anatomy & Layout Control)

The ideal middle ground. Keeps the composition clean, fixes basic hand/finger issues, and blocks multi-panel layout splits or monochrome leaks.

worst quality, low quality, lowres, score_1, score_2, score_3, bad hands, missing fingers, extra fingers, long fingers, monochrome, greyscale, 2koma, 4koma

🔴 Heavy Clean (Maximum Filtering)

Use this if you are getting stubborn artifacts, text/watermarks from background datasets, or scanning noise, while keeping composition and anatomy locked down.

worst quality, low quality, lowres, score_1, score_2, score_3, bad hands, missing fingers, extra fingers, long fingers, blurry, jpeg artifacts, watermark, artist name, signature, text, username, scan artifacts, monochrome, greyscale, 2koma, 4koma