BLIP

BLIP

Runs Salesforce's BLIP vision and language model in a Gradio page, with the training scripts for captioning, retrieval and question answering in the repository. For researchers and developers working with images and text.

Made by Salesforce 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 Gradio app from Salesforce runs BLIP, a model that works with both pictures and text. The repository holds the project code, with scripts named for captioning, visual question answering, retrieval and NLVR, plus a sample image. It names the Salesforce/BLIP model and two BERT models from Google. You use it in a browser, or you run a copy on your own computer.

Who it suits: It suits researchers and developers who work with images and text, and students who want to see a vision and language model at work.

Industry | Who uses it | How they use it | Example
Research | Multimodal researchers | Try BLIP on their own pictures | Open the Space and use the sample image
Education | Machine learning students | See what a vision and language model can do | Read train_caption.py next to app.py
Software | Python developers | Study how the BLIP code and the demo fit together | Download the Space and open app.py

Skip it if you need a light install, because the repository holds 39 files and the model may need a large download.

Where to use it:

• A web browser: go to the Space on Hugging Face and try it with no install
• Windows, macOS or Linux: run a copy with Python

Use case | Who | What they get back
Try BLIP on a picture | Students | The model's output for the image they give
Compare with your images | Researchers | A quick look at how it handles different pictures
Read the BLIP code | Developers | Training and evaluation scripts for several tasks

How to set it up: You need a few things in place first.

• Python with pip, on Windows, macOS or Linux
• An internet connection to fetch the files from Hugging Face
• Spare disk space, since the app names three models and the first run may need extra downloads

1. Install the Hugging Face hub package.
2. Download the Space into a local folder.
3. Move into that folder.
4. Install Gradio 3.0.17 and the packages listed in requirements.txt.
5. Start the app.

pip install -U huggingface_hub
hf download Salesforce/BLIP --repo-type space --local-dir BLIP
cd BLIP
pip install gradio==3.0.17
pip install -r requirements.txt
python app.py

How to use it: Online, use the Gradio interface the Space page shows. On your own computer, run the main file from the project folder and open the page it serves in your browser.

python app.py

Example: You run python app.py inside the project folder, open the page and give it starrynight.jpeg from the repository. The page returns what the model makes of that picture.

For newcomers and experts: If this is new to you, start online. Go to the Space and try the sample picture before you install anything. A vision and language model reads a picture and works with words about it.

If you know the field, note that the repository bundles scripts for pretraining, captioning, retrieval, visual question answering, NLVR and nocaps evaluation. It pins Gradio 3.0.17, an old release, so a fresh environment may need that exact version.

Open the Space or run app.py
The Gradio page loads
You give it a picture
The BLIP model processes it
The result appears on the page

Good to know: The source gives no accuracy figures, so test the output on your own pictures before you rely on it.

• License: bsd-3-clause
• Cost to run: none named, no API key, account or paid service
• Data it reaches: the pictures you give it, the files in the repository and the models it names

On Hugging Face, 13 November 2026

• Likes: 431
• Built with: gradio
• License: bsd-3-clause

Links

• Space: https://huggingface.co/spaces/Salesforce/BLIP
• Files: https://huggingface.co/spaces/Salesforce/BLIP/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.