Krea2 Anime Style Prompt Text

Every style-library node has a sibling nobody talks about until they need it, and this is that sibling. Krea2AnimeStylePromptText runs the exact same anime style selection as its CLIP-encoding cousin - same 152 styles, same checkbox panel, same optional translate/refine - but instead of encoding anything, it hands you the finished prompt strings. You get positive_text and negative_text out, ready to paste, inspect, or feed into whatever else you're building.

Why you'd take the text over the conditioning

Three honest reasons. First, to see what you're actually sending. The checkbox panel shows each style's English sentence, but seeing the full assembled prompt - your words, plus whatever translation or refinement rewrote, plus the selected style sentences comma-joined - is how you learn whether a style you ticked is doing anything at all. Second, reuse: because it outputs plain strings, one style selection can drive several CLIP encoders, so you can feed two different checkpoints from the same ticked boxes without re-picking styles. Third, subgraphs: the README is explicit that this node exists for workflows that want to pass the processed prompt text into another node.

How it works

The mechanism is shared with its sibling - same pipeline, positive prompt in, optional translation to English, optional LLM refinement, then selected style descriptions appended, with your negative prompt only passing through translation. The difference is where it stops. No CLIP encoding happens, which is why the clip input is optional: you only need one when translate_prompts or refine_prompt is on. With both flags off, this node runs with no CLIP connected at all - handy if you're on a setup where you'd rather not load a big LLM text encoder just to shuffle strings around.

The inputs that matter:

• positive_text and negative_text - your prompts, plain and simple.
• style_selection - the same grouped checkbox panel, and it still accepts manual comma/newline-separated style IDs or zh/ja/romaji aliases as plain text.
• translate_prompts / refine_prompt, each with a max-length cap (1024) - both need a text-generation-capable CLIP on clip, or you'll hit the same RuntimeError as the sibling: "Use a text-generation capable CLIP model, or turn translate_prompts/refine_prompt off."

One nice detail from the source: generation runs greedy (do_sample=False), so translation and refinement are deterministic - same input, same output, no surprise drift between runs.

Installing it

Identical to its sibling, since they ship in one pack. ComfyUI Manager, search "ComfyUI-Krea2-AnimeStyle", or:

cd ComfyUI/custom_nodes
git clone https://github.com/eastmoe/ComfyUI-Krea2-AnimeStyle

Restart, and it appears under eastmoe -> Comfy-Krea2-AnimeStyle. No requirements.txt, no model downloads - the style data is bundled in the repo.

Gotchas

The clip input being optional is exactly as fragile as it sounds: it's optional until you enable translation or refinement, and then it's mandatory and must be a generation-capable CLIP. Also worth carrying over from the sibling: if you feed that negative_text into a CFG-1 distilled model like Krea 2 Turbo, it does nothing at the sampler - the node doesn't care, it's just text, but the CFG-1 reality still applies downstream.

Honest take: it's a utility node, not a flashy one. But if you've ever wanted to eyeball what a style library is actually injecting into your prompt before committing to a render, this is the one you reach for.