SatyrRemix – Anima Series

SatyrRemix – Anima is the first Anima model series from SatyrRemix, trained with my previously published Illustrious works as part of its style foundation. Compared with my older Illustrious models, the series has a stronger style identity, more iconic coloring, and a cleaner fantasy illustration finish.

It is best suited for vivid anime/fantasy artwork, character portraits, kemono and furry designs, monster girls, mythic beings, ornate armor, glowing magical effects, and colorful cinematic scenes.

Key style traits include saturated cyan, gold, purple, teal, and sunset palettes, glossy cel-shaded rendering, clean anime linework, expressive eyes, soft but readable faces, shiny metal and water highlights, and decorative fantasy details such as horns, wings, tails, armor, magical weapons, and glowing elements.

Versions

SatyrRemix Anima v1

Establishes the series’ vivid fantasy-illustration foundation, with strong colors, glossy materials, and a pronounced style identity.

SatyrRemix Anima V2 Lite

Retains the iconic SatyrRemix colors, clean linework, and glossy rendering while applying the style more gently. It is designed to be more stable, creative, and flexible when interpreting prompts.

SatyrRemix Anima 2.9B

Brings the series to the expanded Anima 2.9B architecture. In my testing, it follows detailed prompts more consistently and provides a better balance between character presentation, background composition, and the SatyrRemix style.

The expanded base brings broader anime knowledge and coverage of characters and concepts used in Danbooru-style prompting. Its upstream training includes an additional 1.7 million anime/illustration samples, with a reported knowledge cutoff of July 2026. Recognition still varies by character and concept.

This version was tuned and selected to be used alongside a Turbo LoRA. The recommended pairing is Anima-2.9B_Turbo-BF16.safetensors at strength 0.7.

The standard SatyrRemix_Anima_29b checkpoint does not include a baked-in Turbo LoRA. Please load it separately when using the recommended low-step settings below.

Recommended Settings — Anima 2.9B

• Turbo LoRA: Anima-2.9B_Turbo-BF16.safetensors

• Turbo strength: 1

• Sampler: Euler

• Scheduler: SGM Uniform

• Steps: 14–16

• CFG: 1.4

• Shift: 3 to 5

• Starting portrait resolution: 864 × 1296

Start with 16 steps, then try 14 for faster generation. Adjust LoRA strength to suit your prompt and preferred balance of style and detail.

Using Turbo with ComfyUI

Use an Anima workflow that supports the full 40-block Anima 2.9B model, together with the appropriate text encoder and VAE.

1. Place Anima-2.9B_Turbo-BF16.safetensors in ComfyUI/models/loras.

2. Add a Load LoRA (Model Only) node between the diffusion-model loader and the sampling portion of your workflow.

3. Select the LoRA and set strength_model to 0.7.

4. Use euler / sgm_uniform, 14–16 steps, CFG 1.4, and set your Anima workflow’s sampling shift to 3 to 5.

ComfyUI applies the LoRA through nodes. You do not need to include the Forge <lora:...> syntax in the prompt.

Results may differ between Forge Neo and ComfyUI, even with matching settings and seeds. Treat these settings as a starting point for your workflow.

Recommended Settings — v1 and V2 Lite

For the earlier releases, the original starting recommendation remains:

• Recommended workflow: use a Turbo LoRA compatible with the original Anima model.

• Sampler: Euler or ER SDE

• Scheduler: Simple

• Steps: 16

• CFG: 1

The Anima-2.9B_Turbo-BF16 recommendation above is for the 2.9B release.

Prompting

Positive prompt starter:

masterpiece, best quality, very aesthetic, original, highres,

Negative prompt starter:

worst quality, low quality, blurry, jpeg artifacts, censored, watermark, signature, text, malformed hands, extra fingers, bad anatomy, bad perspective,

For 2.9B, combine Danbooru-style tags with clear descriptions of the pose, clothing, framing, lighting, and background. When requesting a named character, include the series name and any important appearance details.

Detailed prompts generally give you more control than a character name followed by quality tags alone.

Notes

The series’ strongest points are color identity, expressive faces, fantasy polish, glossy material rendering, and character-focused illustration.

Complex hands, weapons, animal bodies, and multi-limb creature anatomy may still benefit from clearer prompting, references, or compatible ControlNet workflows.

The recommended low-step settings assume a Turbo LoRA is active. Generation without Turbo requires different settings.