🌸 WDanbooru Negative Preset

The WDanbooru Negative Preset is the simplest node in the pack, and honestly the one you'll wire into every workflow once it's there. It's a dropdown of prebuilt negative prompts tuned for Anima and the other Danbooru-trained models, plus a field for your own additions. No fetching, no API, no model download - it's a string lookup table with an optional append.

If you've ever kept the same negative string pasted into every workflow you own, you already know why this exists.

How it works

Select a preset, and it outputs one string; whatever you type into extra_negative gets appended to it. That's the entire mechanism. What makes it worth installing is that the presets encode the right kind of negative for these models. anima_default is literally Anima's card-recommended base - worst quality, low quality, score_1, score_2, score_3, artist name - which matters because the old forty-keyword SD1.5 boilerplate mostly doesn't survive an A/B test on modern models. These are lean and model-specific instead of cargo-culted.

Picking a preset

Seven choices, and they're self-describing once you know what each targets:

• anima_default - the card's base. Fine for most runs.
• anima_detailed - adds bad anatomy, bad hands, extra fingers, watermarks, text. The all-rounder most people end up on.
• anima_portrait - suppresses full body, wide shot, cloned faces; for close-ups.
• anima_multi_character - kills the classic body-fusion and clone-face artifacts when you're generating several characters.
• anima_sfw_strict - slaps on nude/nudity/nsfw terms for borderline source material.
• quality_only - just worst quality, low quality.
• empty - blank, so only your extra_negative text is used.

One tip the presets don't include: the community's single most-upvoted Anima negative is deviantart - it's a literal dataset tag from training, so negating it steers away from that whole slice of training data. Add it to extra_negative and it rides along on whatever preset you picked.

Output

A single negative_prompt STRING - wire it straight into your CLIP Text Encode negative input. Since it's just text, you can also feed it anywhere else that takes a string, which keeps one consistent negative across an entire workflow.

Installing it

Part of the WDanbooru_Injector pack, so: ComfyUI Manager, search WDanbooru_Injector, install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/Winnougan/WDanbooru_Injector.git

Restart, and all three nodes show up under WDanbooru. Only dependency is requests, which ComfyUI already has.

One thing to keep in mind

Negatives only bite when CFG is above 1 - at CFG 1 the unconditional pass isn't even computed, so the field is silently ignored. That's not a bug in this node; it's just how guidance works. Anima runs real CFG, so these presets do their job there. If you're ever on a distilled model and "nothing changes," check the CFG before you blame the preset.