anima / comfy / createImage

Anima is the first anime model since Illustrious that made people actually switch. It's not another SDXL fine-tune - it's a 2B diffusion transformer on NVIDIA's Cosmos-Predict2, and it fixes the one thing every Danbooru-trained anime model could never fix: prompt comprehension. Civitai Image Gen (anima / comfy / createImage) runs it on the cloud, so you get the prompt understanding without the "surprisingly slow for 2B" reality of running it locally.

What it is

The text-to-image node for the anima ecosystem through the Comfy engine. Required input is prompt; everything else is optional. The KB's anima essay explains why the model matters: Illustrious and NoobAI were trained so deep on Danbooru tags that their text encoder's general knowledge got overwritten, while Anima's Qwen3-0.6B encoder handles tags and natural language in one prompt, and multi-character scenes actually hold together.

Inputs that matter

• prompt - write naturally, or tag, or mix - Anima understands both. The KB notes the deviantart and ye-pop dataset tags act as steerable tokens if you want that flavor.
• diffuser_model - an optional CIVITAI_AIR socket if you want to pin a specific Anima checkpoint or fine-tune; leave unwired for the default.
• width / height (64–2048, default 1024), steps (default 30), cfg_scale (default 4), seed, quantity (1–12).
• sampler / scheduler - the standard ComfyUI-style menus; the KB's local testing found euler/simple works well.
• loras - wire the pack's LoRA Selector. Anima's LoRA ecosystem is its other big draw.
• control_nets - yes, this node has it. Notable, because the KB flags that Anima's local ecosystem was slow to get ControlNet; the cloud version skips that whole wait.

Outputs: images (native IMAGE), errors, workflow_id, raw_json.

Install & gotchas

Standard pack install - ComfyUI Manager (search Civitai Comfy Nodes), comfy node registry-install civitai-comfy-nodes, or git clone + pip install -r requirements.txt. Auth via CIVITAI_API_TOKEN, OAuth, or a wired Civitai Auth node.

Worth knowing:

1. The license is the real gotcha, not the node. Anima's CircleStone Labs license is non-commercial and has a clause (3(c)) the community found genuinely contentious - broadly, commercial partners get rights to your derivatives. Running it here doesn't change that. If your output is commercial, check the license before you spend Buzz on it.
2. "Fast on the cloud" is the point. Locally, Anima is slow for its size - the KB's timings show ~10s on a 5090 and over a minute on a 4060. The cloud removes that; you pay per call instead.
3. Metered, per image. quantity up to 12 means up to 12 billed generations. raw_json shows the cost if you want to verify.

One honest take: if you're in the anime space and the SDXL tag-soup ceiling has been grinding on you, Anima's prompt comprehension is a genuine unlock - and this node is the lowest-friction way to try it before you commit to the local install (and its licensing) for real.