MOSS-Image Character & Style Catalog

MOSS-Image Character & Style Catalog

Browse and search 34,680 generated character and style images, then copy each one's prompt and tags. A read-only viewer for image makers that needs no Python or GPU.

Made by hide1 on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: This is a read-only catalog of 34,680 generated images, made of 34,366 characters and 314 styles. You can search it, filter by eight categories, sort it, enlarge an image and copy its prompt and tags. Nothing in it is edited or uploaded from the viewer.

Who it suits: It suits people who make images and want ready prompts, and people who check how well a model draws named characters and styles.

Industry | Who uses it | How they use it | Example
Image creation | Artists and prompt writers | Browse characters and styles, then copy a prompt for their own generation | Copying the prompt of a film character they like
AI research | Model evaluators | Look through the samples for wrong identities and weak style matches | Spotting a character that does not match its label
Game and story design | Designers looking for looks | Search the catalog for character and style ideas | Filtering to TV series characters

You can skip it if you want only checked, correct images, because nothing here was selected by pass or fail.

Where to use it: You open it in a web browser on its Hugging Face Space. A separate local application exists for work the viewer cannot do.

• The Space in your browser: search, filter, sort, enlarge and copy.
• The local application, from the project's GitHub page and release download, for adding images, editing and comparison judging.

Use case | Who | What they get back
Browse and search | Anyone | Matching characters and styles from the catalog
Copy a prompt | Image makers | The prompt or tags for a card, including a revised full-scene prompt
Check accuracy | Evaluators | Examples of wrong identities and weak style matches

How to set it up: You need no Python, no GPU and no login. Images load only when they are needed.

1. Open the Space in your browser.
2. Give the images a moment to load as you browse.
3. For adding images, editing or comparison judging, get the local application from the project's GitHub page and its v0.1.2 release.

How to use it: Search for a name, pick a category, sort the list, then enlarge an image you like. Open its card to copy the prompt or tags. The details panel holds the original generation prompt.

Tool | What it does
Search | Finds characters and styles in the catalog
Category filters | Eight style categories, plus separate Films and TV Series filters
Sort | Orders the catalog
Enlarge | Shows an image larger
Copy | Copies a card's prompt or tags

Example: You open the Films filter, search for a character you know and enlarge the image. You then open its card and copy the revised full-scene prompt, ready to use in a new generation.

For newcomers and experts: Think of it as a photo gallery where every picture comes with the text that describes it. A good first try is to search a character you know and read its prompt.

If you work with prompts, note how the text was cleaned on 2026-10-31. Every one of the 34,249 entries went through a rule-based text check. Tied-hairstyle tags became plain descriptions of hair. Where the appearance said android or robot, the conflicting human-species label was removed and gender kept. Full-scene prompts now use a neutral whole-character framing in place of the old ears, tail and feet line. The pictures were not regenerated.

Prompt text revised by rules on 2026-10-31
432 character samples added on 2026-11-10
Static viewer loads images on demand
You search, filter and copy prompts

Good to know:

• The catalog is uncurated. It holds wrong character identities and weak style matches too, and being listed does not prove the model trained on that character.
• The newest samples came from an SFT INT8 ConvRot model with a LoRA at 1.0. They ran 8 steps at CFG 1.0 with seed 8961184, at 896×1184, saved as WebP quality 70.
• Revised prompts are for new generations. They are not the text that produced the picture you see.
• No controlled test has shown fewer unwanted tails.
• The 2026-11-10 update added 432 characters and removed one older entry whose actor was unspecified.
• Browser display preferences stay in your own browser. The viewer holds no login or write token.
• The code uses the MIT license. Generated images and the rights of third parties are covered separately in a notice file in the project.

On Hugging Face, 6 November 2026

• Likes: 59
• Built with: static
• License: mit

Links

• Space: https://huggingface.co/spaces/hide1/MOSS-Image-Character-Catalog
• Files: https://huggingface.co/spaces/hide1/MOSS-Image-Character-Catalog/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.