Web App: Booru Prompt Gallery - Get Prompts quickly! (V10.0)
APP LINK: https://booru-prompt-gallery.vercel.app/
ALT LINK: https://booru-prompt-gallery.netlify.app/
CHANGELOG: https://civitai.com/articles/17747
GITHUB: https://github.com/Mexes-GM/booru-prompt-gallery
TL;DR: A web tool to quickly get clean prompts from Danbooru, Gelbooru, e621, Aibooru and Rule34. It removes irrelevant tags, sorts every tag into categories (Appearance, Clothing, Equipment, Pose, Scenery, Creature...), merges redundant tags, avoids contradictions, and lets you mix posts together or build a whole batch of prompts from a single one. Prompts are designed for Illustrious, Pony, SDXL and Anima, and you can also convert them to natural language with AI.
Hello! The last main article was written back in V8.0, and since then the app has changed so much that half of what was explained there doesn't even exist anymore. V10 is the biggest update so far, so I decided to remake the article one more time, explain everything from scratch and leave this as the new main article.
What is this?
Booru Prompt Gallery, as its name suggests, is a gallery... of prompts. It takes posts from digital art websites (like Danbooru), extracts the tags that describe them, cleans them up and sorts them so they're ready to paste into your generator. It's extremely useful for people who train LoRAs, test checkpoints, or just want good prompts without writing them by hand.
Why did I make it?
As a LoRA and checkpoint trainer, the most time-consuming part for me was testing the model and creating varied, high-quality examples for its release. So, looking to speed up that part, I made this page. At first it was just a personal tool, but it turned out to be useful to a lot more people, so here we are.
Feature Breakdown
From here on down, I'll explain "what it does and how to use it" for every tool in the app. I tried to add an image to almost everything, so the article looks much longer than it is. It's not as much text as it seems, trust me! (ok, maybe it is).
Basic Operation
Type what you want to see in the search bar. For example, if you want examples of Frieren, just type "frieren" and you'll only get posts with her. This works for any booru tag: characters, clothing, poses, places, etc. If you're not sure how a tag is written, the autocomplete will help you (it knows ~145,000 Danbooru tags and their aliases, and shows you what type of tag each one is).
After that, just look for a card you like and press Copy. That's it, the prompt is in your clipboard.
Note: you won't get the exact image, but you'll get everything that describes it. How close it gets depends on how well tagged the post is.
If you click a card, it expands so you can read the whole prompt.
If you only want part of the prompt, the arrow next to Copy lets you copy just one category, like the clothing or the pose. The number next to each one tells you how many tags it has.
The Search Panel
- Site selector: choose where the posts come from. Danbooru, Gelbooru, e621 (furry), Aibooru (AI-generated art) and Rule34. Generally I recommend Danbooru, since its tagging is the best and the most suitable for Illustrious.
- Search bar: here you type what you want to see. Right under it you can see exactly what's being sent to the site, and how many tag slots you're using. Each site has its own limits (Danbooru only allows 2 tags for anonymous users, for example), and the app will tell you when you go over.
- NSFW switch: the typical content shield. It's off by default, so you only get safe content unless you turn it on.
- Shuffle: gets random results instead of always showing the latest posts. Very useful if you're tired of seeing the same things in the same order.
- Card size: the magnifying glass buttons change the cards between Small, Medium and Large.
Search Filters
These options change which posts you get from the site.
- Minimum tags per post: only shows posts with at least this many tags. More tags = more detailed prompts. I recommend 25+.
- Post quality: filters by score. From our testing, posts with higher scores tend to be much better tagged, so this is a quick way to get cleaner prompts.
- Only known characters: hides characters with very few posts. Useful if you want characters your model is likely to know; an obscure character with 20 posts probably won't come out right.
- Blacklist: tags you never want to see in your results.
Customize Prompt
These options don't change which posts you get, they change the prompt you copy.
- Add to every prompt: adds your tags at the start of every prompt. Perfect for LoRA trigger words or quality tags. You can save them as Tag Presets (the save icon) and load them later, which is very useful if you work with several LoRAs.
- Skip added tags that don't fit the image: if one of your added tags contradicts the post, it gets skipped. For example, no face tags when the post is a back view.
- Remove from every prompt: removes tags from all prompts. For example "solo" or "white background".
- Include character names: adds or removes the character tags.
- Merge redundant tags: if the prompt has "hair, long hair, white hair", this combines them into "long white hair". Less redundancy, fewer tokens.
- Add my search tags: if you searched for something and the post doesn't have that tag, it gets added anyway.
- Start with @artist: adds the post's artist as "@artist" at the start of the prompt, to replicate their style. Only for Anima checkpoints.
- Global tag weights: applies the weights you set on tags to every card (more on that below).
Background Options
Keep the original background, remove it, replace it with your own tags, or randomize it:
- Simple Random gives each card a simple color/gradient background that won't clash with the image.
- Detailed Random swaps in a full scene (like "outdoors, beach, ocean, day").
Great for testing different scenery while keeping the character, pose and clothing intact.
Find & Replace Rules
Your own rules for the prompt. Replace swaps one tag for another, and Append adds tags after one. For example: every time a prompt has "elf", add "pointy ears, long pointy ears".
How do tag weights work?
All tags are clickable. Clicking one opens a small panel where you can raise or lower its weight, copy the tag or search it directly. If you set it to 1.3, for example, that tag will be copied as (tag:1.3). You can also right-click any tag to copy it.
If you turn on Global tag weights, the weights you set are saved and applied to every card that has that tag, and you can manage all of them from the panel.
The Card Menu
Hovering over a card shows a heart to save it to your favorites (the arrow next to it lets you pick a folder) and a "..." menu with more options:
- Save artist: saves the artist to a dedicated folder in your favorites.
- Convert to natural language: sends the prompt to AI Convert.
- Make pack: starts a Pack with this card as the base (explained below).
- Download image: useful if you need the image for ControlNet or IP-Adapter.
- View original post: opens the post on its site.
Modes
The app has three ways of working with prompts: Browse (the normal one, copy prompts one by one), Merge and Pack.
1. Merge
In Merge mode each card shows its categories, and you can pick which ones you want from each post. Character from one card, outfit from another, background from a third... and all of it ends up in a single prompt, without duplicates. You can pick more than one category per card and combine as many cards as you want. If there's a tag you don't like, click it and it's gone.
- Random fills the merge with random picks.
- To Variations turns it into a wildcard prompt { option A | option B }) so every generation picks a different combination.
2. Pack Mode ✨
This is the big new feature of V10. Pick a base (a card from the gallery or your own prompt, for example your OC) and Pack Mode builds a whole batch of prompts around it: same character, different outfits, poses, expressions and backgrounds. Perfect for testing a character LoRA, or making a dataset.
Then you choose where the variations come from: SFW or NSFW, which site, and which posts to sample from.
In the builder:
- Each category can be set to Keep (use the base's tags), Vary (take tags from the collected posts) or Off.
- Choose how many prompts you want, how different they should be with the Variety slider, and the minimum number of tags per prompt.
- The "Always add" field goes first in every prompt, exactly as you typed it (LoRA triggers, quality tags, etc).
- Don't like one of the results? Re-roll just that one without touching the rest. When you're happy, Copy all.
The more you use it, the more it learns what you keep. It also avoids silly mixes like "white background" with the character standing in a classroom, or two outfits at once.
Tip: you can also start a pack from any card with "..." > Make pack.
AI Convert
Converts booru-style prompts into natural language using Vision and LLM models. Useful for models like Anima that understand natural language better. Every user gets 10 free requests per day with free-tier models that work decently well, and if you want unlimited use, you can plug in your own API key from the Settings button in the panel.
To use it, press the star button at the bottom-right of the page (every card's button turns into "Convert"), or use "Convert to natural language" from a card's "..." menu.
Favorites and History
Favorites saves the posts you like. You can organize them into folders, and saved artists get their own folder too. If you sign in, they sync between devices.
History shows every prompt you've copied, from any site, so you can copy it again later.
Tools
Import & Clean
Paste your own prompt (or drop an image with metadata) and the app cleans it, strips weights and LoRA tags, and sorts it by category so you can choose what to keep.
Quick Teach
Helps improve the tag categories. It deserves its own section, so keep reading.
Tag Categories and Quick Teach
Sorting tags into categories couldn't be done by magic. Before, the app used 5 broad categories, and a lot of common stuff simply had nowhere to go: swords, phones, umbrellas, food, cats, art styles... all of it ended up dumped into "other".
Now there are 7 categories (Appearance, Clothing, Equipment, Pose, Scenery, Creature and Other), split into 33 subcategories like hair, eyes, top, bottom, footwear, weapon, handheld, expression, camera angle, setting, props and more. To classify all ~145,000 Danbooru tags I used Jev, an AI model from TypeSafe made for this kind of judgment call. This is what lets Merge, copy by category and Pack Mode know what each tag actually is.
The tags Jev wasn't sure about (is "flower" in her hair, in her hand, or in the background? is "blush" an expression or a physical thing?) were set aside for humans to decide, and that's what Quick Teach is for:
- It shows you those tags one by one with Jev's guess already picked.
- Press Enter to accept it, or choose the right category and subcategory with the number keys.
- Every answer is saved right away, and you can undo if you misclick.
Thank you very much to everyone who decides to collaborate!
Account and Settings
You don't need an account to use the app. If you sign in (just your email, you'll get a magic link, no password), your favorites and preferences are saved and synced between devices.
The "..." menu at the top-right lets you switch between light and dark theme, replay the tour, send me feedback, and check the service status in case something isn't working.
Mobile
The app works on mobile too. It used to be clunky, I must admit, but it got a big overhaul and now it should feel a lot smoother. The panels open as bottom sheets, so everything stays within thumb reach.
Open Source
The project is open source, so if the site is ever down (it runs on free hosting plans, so sometimes resources get exhausted), you can also run it locally. Instructions are on GitHub: https://github.com/Mexes-GM/booru-prompt-gallery
FAQ
Found a bug?
Don't hesitate to tell me — I'm happy to fix them all.
Have feature suggestions?
Absolutely! Your ideas are welcome. A lot of the features in this article were suggested by you.
Is the site down?
The app runs on free hosting plans, so sometimes resources get exhausted. If the main link is down, try the alt link. I apologize in advance if that happens.
Want to support me?
Thank you so much! You can [Buy Me a Coffee](https://buymeacoffee.com/Mexes), or send Buzz. It also helps a lot if you:
- Leave a comment and a rating.
- Use my models on the platforms linked in my profile (especially TensorArt and SeaArt).
That's it!
Stay hydrated and don't forget to blink.
