Laya Vision

Laya Vision

Answers yes/no, multiple-choice and rubric-scored questions about an image in a single pass, with no text generation. It runs a small 201M checkpoint on a CPU and suits vision researchers and developers.

Made by thaitea 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: Laya Vision answers structured questions about an image. It handles calibrated yes/no questions, multiple-choice questions and rubric-graded scores. Each answer takes one pass through the model and nothing is written out. A small vision-language checkpoint does the work on a CPU.

Who it suits: Vision researchers and developers who want a fast, calibrated check on an image instead of free-form text. It also suits people studying small vision-language models.

Title: Who asks structured questions about images
Industry | Who uses it | How they use it | Example
Research | Vision-language researchers | Test a small model's calibrated answers on images | Ask a yes/no question and read the calibrated result
Software | Developers | Add a quick image check to a pipeline | Check whether an image meets a yes/no condition
Content review | Reviewers | Score images against a rubric | Grade a picture with the score mode
Education | Students | See how a pruned model answers questions | Pick the right option for an image with choice

If you want long written descriptions of an image, you can skip it, since the model does no text generation.

Where to use it:

• A web browser: the Space runs on Hugging Face with nothing to install
• Your own computer, running Python 3.12 on Windows, macOS or Linux

Use case | Who | What they get back
Ask a yes/no question about an image | Developers | A calibrated answer
Choose between options for an image | Researchers | The option the model picks
Score an image against a rubric | Reviewers | A rubric-graded score

How to set it up:

• Python 3.12 and pip
• The Python packages in requirements.txt

1. Install the Hugging Face hub package and download the Space into a folder.
2. Move into that folder.
3. Install Gradio 6.28.0, then the packages from requirements.txt.
4. Start the app with Python.

pip install -U huggingface_hub
hf download thaitea/laya-vision-demo --repo-type space --local-dir laya-vision-demo
cd laya-vision-demo
pip install gradio==6.28.0
pip install -r requirements.txt
python app.py

How to use it: Open the Space in your browser, or start the local copy with the command below and open the page Gradio serves. Give it an image and choose one of three question types, then ask your question.

python app.py

Tool | What it does
noul | Answers a calibrated yes/no question about an image
choice | Answers a multiple-choice question about an image
score | Grades an image against a rubric

Example: You give it a photo of a cat and use the yes/no mode to ask whether a cat is in the picture. It returns a calibrated yes or no from a single pass, without writing any text.

For newcomers and experts: If you are new to this, open the Space and try the example photo with the yes/no mode first. Then try the choice and score modes on the same picture.

If you know the field, the Space runs thaitea/laya-vision at a pinned revision, a 201M checkpoint made by cutting SmolVLM-256M down to 20 of its 30 language layers. Training drew on The Cauldron, on frames from games and on four datasets with rubric scores. The revision is pinned in app.py, so results stay repeatable. The trade is a small model that answers fast on CPU but cannot write.

You give it an image and a question
One forward pass runs on CPU
A calibrated answer or score returns

Good to know:

• Hardware: the Space runs on cpu-basic at Hugging Face
• Pinned model: the revision is fixed in app.py through LAYA_REVISION
• Example photo: it is credited to Alvesgaspar under CC BY-SA 3.0
• License: cc-by-nc-sa-4
• Cost to run: none named
• Not for: writing text, since there is no text generation, and commercial use, since the license is non-commercial

On Hugging Face, 13 November 2026

• Likes: 10
• Built with: gradio
• License: cc-by-nc-sa-4.0

Links

• Space: https://huggingface.co/spaces/thaitea/laya-vision-demo
• Files: https://huggingface.co/spaces/thaitea/laya-vision-demo/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.