Anima Prompt T8

If you've moved to Anima, you already know the workflow pain: the model wants its prompt in a specific shape - subject, negatives, and a style/artist layer that quietly does half the work. Anima Prompt T8 is the centerpiece of the comfyui-anima-t8 pack, and it exists to give that prompt one home. It's not doing anything magical, which is kind of the point. It's three labeled text boxes that hand ComfyUI clean strings, so you stop rebuilding the same prompt inside CLIPTextEncode every time.

How it works. The node takes three multiline inputs - positive, negative, and style - plus an optional merge_separator (defaults to ", "). It returns four strings: POSITIVE, NEGATIVE, STYLE, and MERGED_POSITIVE, where the last one is simply your positive prompt joined with the style text. That split is the whole design: keep your subject words and your artist tags in different boxes, then decide at the end whether they get encoded together or separately.

The defaults are worth reading because they tell you what this pack assumes. The negative box ships pre-filled with an Anima-friendly negative - lazyneg, bad anatomy, bad hands, watermark, all the usual suspects - which most people just keep. The positive box opens with score_9,score_8,score_7,1girl, masterpiece, best quality, ultra detailed. That's the Pony-compatible prefix, and it's where I'd push back a little. Anima genuinely understands Pony-style score_x tags - it was trained with them and the model card itself recommends score_7 - but the community consensus is that score_8 and score_9 push output toward that generic, polished "furry classifier" look. If your generations all come out samey, cut the prefix down to score_7 or drop it entirely, especially once you're already naming an artist. It's Pony vocabulary, not a law of physics.

What you actually set. Just the three boxes, honestly. positive is your subject and scene, negative your cleanups, style where the artist tags land. Wire POSITIVE into the positive CLIPTextEncode and NEGATIVE into the negative one. Want the style baked into the same conditioning? Use MERGED_POSITIVE for the positive encode instead. The STYLE output exists so you can see the style layer on its own, and it's the port the pack's other nodes - Anima Artist Style T8, the Saved Prompt Loader - naturally feed into. You'll find it under Add Node → Anima/T8.

Installing. The pack is one clone away:

cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/comfyui-anima-t8
cd comfyui-anima-t8
pip install -r requirements.txt

Restart ComfyUI, or search comfyui-anima-t8 in ComfyUI Manager and let it do the work. The only real dependency in requirements.txt is requests; Pillow, numpy, and torch come with ComfyUI. No model download here - this is a prompt node, not a checkpoint. You still need the Anima model itself from HuggingFace.

Where people get burned. The classic one: you filled in style, wired POSITIVE into your encoder, and your artist tag never shows up in the image. That's not a bug - POSITIVE deliberately excludes the style layer, because the whole point is that you can choose. Wire MERGED_POSITIVE and it'll be there. Second classic: results look "generic polished" and you assume Anima is overrated. It's the score_9 prefix narrowing the output distribution. And if you're running a pure Pony model instead, note these defaults only carry the first three score tags, while Pony's trained convention wants the full score_9, score_8_up, score_7_up string. This node is Anima-shaped first; for Pony it's a decent starting point, not a finished one.