Anima2 for Text to Image

Description:

Anima 2 Preview text-to-image generation with a Qwen 3 0.6B text encoder and AuraFlow-style noise scheduling.

Describe a scene, set your resolution, and hit Run. The workflow generates four images per batch so you can pick the best result without re-running. Defaults are set for portrait-oriented output at 896x1152.

How do you use Anima 2 for text-to-image generation?

Write a positive prompt describing your scene, adjust the resolution if needed, and run. The workflow comes with optimized sampler settings (er_sde, 30 steps, CFG 4) and generates four images per batch. Most of the defaults work out of the box.

Positive Prompt This is where you describe what you want. Anima 2 responds well to detailed scene descriptions with style cues. The example prompt uses phrases like "fantasy anime style, highly detailed, atmospheric" to steer the output. Want a specific look? Stack style descriptors at the end of your prompt. Need more control over composition? Describe spatial relationships ("floats above," "beneath his feet") and the model picks them up.

Negative Prompt Tells the model what to avoid. The default covers the basics: "worst quality, low quality, score_1, score_2, score_3, blurry, jpeg artifacts." You can add specific things you don't want (extra limbs, bad anatomy) but the default handles most cases.

Resolution (Width / Height) Default is 896x1152, which gives you a portrait orientation. Want landscape? Swap the values to 1152x896. Square works too at 1024x1024. The model handles standard SDXL-class resolutions. Going much higher than 1.5 megapixels per image may slow things down or affect quality.

Batch Size Set to 4 by default. Each run gives you four different images from the same prompt. Generating more takes longer but gives you more options. Drop it to 1 if you're iterating fast on a prompt.

CFG (Guidance Scale) Default is 4. This controls how closely the model follows your prompt. Lower values (2 to 3) give the model more creative freedom. Higher values (5 to 7) make it stick closer to your words. Too high and you'll start seeing artifacts and oversaturation. Start with the default and adjust from there.

Steps Set to 30. More steps means more detail but longer generation time. For quick drafts, 20 works fine. For final renders, 30 to 40 gives good results. Going past 50 rarely helps.

Seed Controls randomness. Keep it fixed to reproduce the same image with the same settings. Change it or set it to random to get fresh results each run.

What is Anima 2 Preview good for?

Anima 2 Preview is built for anime and illustration-style image generation. It pairs well with detailed scene prompts and produces images with strong composition and consistent character rendering. Best for character design, concept art, and stylized illustrations.

If you're working on character designs, this is a strong starting point. Describe your character's appearance, pose, and environment. The model handles fantasy and sci-fi themes well based on the example prompts.

For concept art exploration, the batch-of-four default lets you see a range of interpretations from a single prompt. You can iterate on the ones that work and refine from there.

This is a preview model. Expect ongoing improvements. If you need photorealistic output, this isn't the right workflow. Look for Flux or SD 3.5 workflows instead.

FAQ

What resolution should I use with Anima 2 Preview? The default 896x1152 works well for portrait-oriented images. Swap to 1152x896 for landscape. Stick to standard resolutions in the 0.7 to 1.5 megapixel range for best results. The model uses the same latent space as other SDXL-class models.

What CFG value works best for Anima 2? Start with the default of 4. This gives a good balance between prompt adherence and image quality. If your images look washed out, try bumping it to 5 or 6. If they look oversaturated or have artifacts, drop it to 3.

How many steps do I need for Anima 2 text-to-image? 30 steps is the default and works well for most prompts. For quick tests, 20 is enough to see if a prompt direction works. For polished output, stay at 30 to 40. Going beyond 50 gives diminishing returns.

What is the ModelSamplingAuraFlow node in this workflow? It adjusts the noise scheduling for the model with a shift value of 3. This is already configured and you don't need to change it. It helps the sampler produce better results with this specific model architecture.

How do I run Anima 2 text-to-image online? You can run Anima 2 text-to-image online through Floyo. No installation, no setup. Open the workflow in your browser, upload your inputs, and hit run. Free to try.