Anima Beginners Guide
🌸 Anima Beginner's Guide
This guide assumes you've already read the Illustrious Beginner's Guide.
Many of the concepts like samplers, CFG, VAEs, schedulers, LoRAs, and prompting work exactly the same. This guide focuses on what makes Anima different.
🌸 What is Anima?
Anima is an anime-focused diffusion model created by CircleStone Labs in collaboration with Comfy Org. Instead of trying to generate every style imaginable, it concentrates on one thing: producing beautiful anime and manga artwork with excellent prompt understanding.
Unlike many anime checkpoints that rely heavily on tag-based prompting, Anima was designed to understand natural language remarkably well. You can still use traditional Danbooru tags, but you're no longer forced to "speak tag" to get good results.
Think of Illustrious as a highly technical illustrator who wants very specific instructions.
Think of Anima as an artist who understands what you mean.
Both can create fantastic images, but they respond to prompts a little differently.
⚙️ Under the Hood
Some of Anima's technology is different from traditional SDXL models.
NVIDIA Cosmos Architecture
Anima is built on NVIDIA's Cosmos Predict-2 Diffusion Transformer, giving it modern diffusion capabilities while remaining relatively lightweight.
Qwen-3 Text Encoder
Instead of relying on a traditional CLIP encoder, Anima uses a 0.6B Qwen-3 text encoder.
This gives it an excellent understanding of:
- Long prompts
- Natural language
- Descriptive sentences
- Character interactions
- Scene composition
This is one of the biggest reasons Anima feels easier to prompt.
Trained on Real Artwork
The model was trained on millions of authentic anime illustrations along with hundreds of thousands of artistic images.
No synthetic AI-generated images were used during training.
Smaller Model Size
One pleasant surprise is the download size.
| Illustrious | Anima |
|---|---|
| ~6-7 GB | ~3-4 GB |
Smaller checkpoints generally load faster, require less VRAM, and often generate images a little more quickly.
🤝 Compatibility with Illustrious
Good news!
Because the two models share similar ancestry, many Illustrious LoRAs work surprisingly well in Anima.
That doesn't mean every LoRA is guaranteed to behave identically, but many character, style, and utility LoRAs transfer with little or no adjustment.
If you've already built a nice LoRA collection for Illustrious, you're probably not starting over.
⚖️ Anima vs. Illustrious
| Anima | Illustrious |
|---|---|
| Understands natural language extremely well | Prefers structured Danbooru tags |
| Flexible prompting style | Very tag-oriented |
| Great at expressive characters | Extremely accurate tag interpretation |
| Easy for beginners | Rewards detailed prompting |
| Excellent artistic flexibility | Excels at highly controlled anime output |
Neither model is "better."
They're simply optimized for different workflows.
⚙️ Recommended Settings
These settings provide an excellent starting point.
| Setting | Recommendation |
|---|---|
| Sampler | DPM++ 2M SDE Karras or DPM++ 2M Karras |
| Steps | 25-35 |
| CFG | 4-7 |
| Resolution | Use the model's native resolution |
As always, don't be afraid to experiment. Small changes can make a noticeable difference.
✍️ Prompting Tips
Anima is much more forgiving than many anime models.
Some recommendations:
- Write naturally.
- Describe what you want instead of trying to tag everything.
- Danbooru tags still work well.
- Keep prompts focused.
- Avoid stuffing every idea into one prompt.
For example, this works surprisingly well:
A cheerful mage standing on a cliff at sunrise, long silver hair blowing in the wind, detailed fantasy clothing, soft lighting, dramatic clouds.
Illustrious would usually benefit from additional structured tags.
Anima often understands the sentence exactly as written.
🎨 Using LoRAs
LoRAs behave much like they do in Illustrious.
Some general guidelines:
- Illustrious LoRAs usually carry over well.
- Character LoRAs perform particularly well.
- Style LoRAs can dramatically change the final artwork.
- Start with a weight between 0.7 and 0.9, then adjust from there.
Remember that stronger isn't always better. Sometimes reducing a LoRA's weight creates a more natural-looking image.
💭 Final Thoughts
If Illustrious feels like giving precise instructions to a professional illustrator, Anima feels like collaborating with an artist who quickly understands your vision.
Both models are capable of producing outstanding artwork.
The biggest difference isn't image quality.
It's how you communicate with the model.
If you enjoy writing descriptive prompts in plain English, you'll probably feel right at home with Anima. If you prefer meticulous control through tags, Illustrious will likely remain your favorite.
The best part?
You don't have to choose. Many creators keep both models installed and simply use whichever one fits the project they're working on.