Animagine Prompt

This one doesn't touch pixels. It doesn't call an API, doesn't need a key, doesn't even need a GPU to run - it's a text node that saves you from retyping "masterpiece, best quality, absurdres, high score, great score, good score" for the thousandth time. If you're generating with Animagine XL 4.0, Cagliostro Lab's anime SDXL finetune, you've probably already seen their official prompting guide: a specific menu of quality tags, score tags, rating tags, and a year tag, meant to be combined in a particular order for best results. This node is that menu, wired up as toggles instead of a wall of text you copy-paste and half-remember.

Why it exists

Animagine XL 4.0, like Illustrious and NoobAI, is trained on Danbooru-style tags rather than natural-language captions - same family, same rules: exact tags for pose, clothing, camera angle beat sentences, and "masterpiece, best quality" only means something because the training data was tagged with it. What's specific to Animagine is the extra layer on top: alongside quality tags it also wants score tags ("high score, great score, good score") and supports explicit rating tags (safe / sensitive / nsfw / explicit) straight out of Danbooru's own content-rating vocabulary, plus an optional year tag to nudge the output toward a particular era's art trends. It's genuinely easy to forget one of these or get the order wrong when you're typing it by hand every time - which is exactly the itch this node scratches. Worth knowing going in: Animagine had its moment in early 2024 as the anime-SDXL model to reach for, before Illustrious largely took that crown. It's still a perfectly capable model and this node still does its job - just don't expect a big active community around either the model or this specific pack.

Inputs and outputs

You write your own scene in positive_prompt and negative_prompt - the node doesn't generate content for you, it decorates what you give it. Everything else is toggle-plus-value pairs:

• use_good_tags (on by default) pairs with quality_good and score_good - dropdowns of pre-built combinations like "masterpiece, best quality, absurdres" and "high score, great score, good score." Leave this on; it's the boilerplate that's actually worth keeping.
• use_bad_tags (off by default) does the same for the negative side via quality_bad/score_bad.
• use_rating + rating lets you pick safe, sensitive, nsfw, or explicit - Danbooru's own rating tokens, appended to steer content rating.
• use_year + year (default 2024) tags the generation toward that year's stylistic conventions.
• use_csv + csv_path + csv_index load a specific row from a CSV of recurring characters, so you're not retyping the same character description every session. The CSV needs exactly three columns - GENDER,CHARACTER,COPYRIGHT - saved as UTF-8.

Two outputs, POSITIVE_PROMPT and NEGATIVE_PROMPT, both plain strings - wire them straight into your two CLIPTextEncode nodes. Nothing here loads a checkpoint; you still need the Animagine XL 4.0 model itself downloaded and loaded separately, this just builds the text that goes into it.

Installing it

ComfyUI Manager: search ComfyUI-Animagine-Prompt, install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/gmorks/ComfyUI-Animagine-Prompt

(The README's own clone command still points at a yourusername placeholder - ignore it and use the real URL above.) Restart ComfyUI and the node shows up under the Animagine-Prompt category. No model downloads triggered by this node itself - it's pure string assembly, so install is fast and nothing heavy comes down.

Where people get burned

Almost everything reported against this node traces back to the CSV feature, per the author's own troubleshooting notes: wrong file path, a CSV missing one of the three required columns, or a file saved in something other than UTF-8 - all three silently produce no character text rather than a clear error, so check those first. Second, watch for conflicting tags: if your own positive_prompt already contains "masterpiece" or a quality phrase and you've also got use_good_tags on, you'll get duplicated boilerplate - harmless, but it eats into how much of your prompt is actually about your scene. And a trailing comma or stray space where your prompt meets the appended tags is a classic source of "why does this look slightly off" - the README calls it out directly as a common issue. Last thing: this is a small, single-author hobby pack (the author has said as much themselves), not a heavily maintained ecosystem staple, so if you hit an edge case, check the GitHub issues before assuming it's documented behavior.