pixai-tagger-v1.0

pixai-tagger-v1.0

Labels anime images from a vocabulary of 30,877 tags in six categories, covering characters, clothing and styles. For people who sort image libraries, build datasets or write captions.

Made by pixai-labs on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: PixAI Tagger v1.0 is an image classification model from pixai-labs that reads an anime picture and returns a score for each tag it finds. Its vocabulary holds 30,877 tags in six categories. It runs on a finetuned SAM3 backbone with 486M parameters and reads images at 1008 by 1008 pixels. The card gives a knowledge cutoff of May 2026.

Who it suits: It suits people who handle many anime images and want tags generated for them, such as collectors, dataset builders and caption writers.

Industry | Who uses it | How they use it | Example
Anime and illustration | Collectors and archivists | Sort a large image library by tag | Finding every picture of one character
AI training | Dataset builders | Prepare labeled images for training | Adding tags to a folder before fine-tuning
Art platforms | Site developers | Add searchable tags to uploads | Suggesting tags when someone posts a picture
Content captioning | Prompt and caption writers | Start captions from a tag list | Turning a tag list into a written description

You can skip it if your pictures are real photographs, since the card describes it as a tagger for anime images.

Where to use it:

• Python with the Transformers library, through its pipeline call
• A setup with PyTorch, torchvision, timm, NumPy and Pillow
• Windows, macOS or Linux with Python, to download the files
• A local folder, if you have already downloaded the model

Use case | Who | What they get back
Library sorting | Collectors | Tags grouped by category for each image
Dataset preparation | Builders | Tag scores to filter and label images
Caption help | Writers | Tag names to turn into captions

How to set it up: You need Python for the download tool, plus the libraries named above. The weights are about 1.9 GB in F32, or roughly 243 MB at 4-bit as an estimate.

• Python on Windows, macOS or Linux
• The huggingface_hub package
• PyTorch, torchvision, Transformers, timm, NumPy and Pillow
• Room for the weights on disk and in memory

1. Install huggingface_hub.
2. Download the model files, or let the pipeline fetch them by name.
3. Install the other libraries, then run the quick start below.

pip install -U huggingface_hub
hf download pixai-labs/pixai-tagger-v1.0

How to use it: The quick start builds a tagging pipeline, opens one image and prints the general tags. Results come back as tag names with scores, grouped by category. The pipeline already applies the recommended thresholds for each category.

from PIL import Image
from transformers import pipeline

model_id = "pixai-labs/pixai-tagger-v1.0"
tagger = pipeline(
    model=model_id,
    image_processor=model_id,
    trust_remote_code=True,
)

image = Image.open("image.jpg")
results = tagger(image)["results"]
print(results["general"])

A higher threshold gives fewer but surer tags, and a lower one lets more guesses through. You can set one category or pass a single number for all of them.

Category | Tags | What it covers
general | 15,043 | Things you can see, such as clothing, poses and objects
character | 8,308 | Named characters
style | 4,917 | Style labels
copyright | 2,460 | Source works and franchises
meta | 145 | Medium and resolution labels
rating | 4 | Content rating labels

Example: You open a picture with Pillow, pass it to the tagger and print the general results. The card's illustrative output shows tags such as 1girl and solo with scores near 0.98 and 0.95.

For newcomers and experts: If you are new to this, the model looks at a picture and lists words that describe it, each paired with a score. A first small try is one image with the default thresholds.

For people who know the field, the authors chose each category's cut-off on their own test set, so use those values as places to start on your own images. The model resizes and pads input without stretching it. The card compares it with seven other taggers on tags they share.

Title: Micro F1 on tags shared with v0.9
v0.9 General: 0.5980
v1.0 General: 0.6660
v0.9 Character: 0.8198
v1.0 Character: 0.9242

You load the tagger with the Transformers pipeline
You pass in an image
The model scores tags in six categories
You keep the tags above each threshold

Good to know: The pipeline loads model code that sits in the repository, which is what the trust_remote_code setting allows, so read it first if you care.

• License: Apache 2.0
• Cost to run: none named
• Data it reaches: the images you pass in on your own machine, plus the downloaded files
• Not for: pictures that are not anime, since the card describes it as an anime tagger

On Hugging Face, 13 November 2026

• Task: image classification
• Library: transformers
• Parameters: 486M
• Weights: about 1.9 GB in F32; about 243 MB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 36,625
• Likes: 165
• License: apache-2.0
• Quantized versions: 7
• Fine-tunes: 3
• Tags: transformers, safetensors, cls_vitdet, feature-extraction, multi-label, anime, image-classification

Links

• Model: https://huggingface.co/pixai-labs/pixai-tagger-v1.0
• Files: https://huggingface.co/pixai-labs/pixai-tagger-v1.0/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download pixai-labs/pixai-tagger-v1.0

Then follow the usage in its model card. Nothing in this kit runs by itself.