Unlock the ability of images with AI Sheets

🧭TL;DR: Hugging Face AI Sheets is an open-source tool for supercharging datasets with AI models, no code required. Now with vision support: extract data from images (receipts, documents), generate visuals from text, and edit images—all in a spreadsheet. Powered by hundreds of open models via Inference Providers.

We’re excited to release a large update to Hugging Face AI Sheets, the open-source tool for constructing, transforming, and enriching data with open AI models. AI Sheets leverages Inference Providers, which implies you need to use hundreds of open models powered by the perfect inference providers on the planet.

The first version of AI Sheets made structuring and enriching textual content a breeze. Now, we’re adding vision to AI Sheets.

Images are in all places—product photos, receipts, screenshots, diagrams, charts, logos. These documents contain structured information waiting to be extracted, analyzed, and transformed. Today, you may finally work with visual content directly in AI Sheets: view images, analyze them, extract information, generate recent ones, and even edit them in real-time —all in the identical workflow.

Your images have stories to inform

Images contain useful information—product catalogs, support tickets, research archives, receipts, documents. Now you may upload images directly or use datasets with images, and use vision models to extract, analyze, and structure the data inside them.

What you may do:

• Describe and categorize images – Generate captions for product photos, classify document types, or tag images by content
• Extract structured data – Pull line items from receipts, data from charts, or text from scanned documents
• Add context and metadata – Mechanically label images with relevant attributes, quality scores, or custom annotations

Identical to text columns, you may iterate on prompts, manually edit outputs, and use thumbs-up to show the model what you wish. Your feedback becomes few-shot examples for higher results.

Example: From receipts to structured expenses

Imagine you are back from a visit with a stack of receipts. Upload them to AI Sheets and create a column with a prompt like: Extract the merchant name, date, total amount, and expense category from this receipt

AI Sheets processes each receipt and provides you a clean table with all the main points extracted. You’ll be able to edit any mistakes, validate good results with thumbs-up, and regenerate to enhance the remainder. Export the ultimate dataset as CSV or Parquet to your expense tracking tool.

Or perhaps you are digitizing handwritten recipes from old family notebooks. Create columns to extract ingredients, cooking time, and cuisine type—turning your personal archive right into a searchable, structured dataset.

Generate and transform text and pictures in the identical flow

Need visuals to your content? AI Sheets can generate and edit images directly in your spreadsheet using AI models, keeping your entire content creation workflow in a single place.
What you may do:

• Generate images from text – Create social media graphics, thumbnails, or illustrations that match your content
• Edit and transform existing images – Modify uploaded images or generated visuals—change styles, add elements, adjust compositions
• Create variations at scale – Generate multiple versions or styles to check what resonates along with your audience
• Construct visual content libraries – Produce consistent branded assets across large content campaigns

Example: Making a content calendar with visuals
Imagine you are planning a month of social media posts about healthy recipes. You might have a spreadsheet with post titles and descriptions, but no images yet.

Create a picture column with a prompt like: Generate an appetizing food photo for: {{title}}. Style: brilliant, overhead shot, natural lighting.

AI Sheets generates a singular image for every post. Not quite right? Create one other column to edit them: Transform the image to have a country picket background and add fresh herbs as garnish.

You’ll be able to iterate on generation and editing prompts and check out different approaches. Your entire content calendar—copy and visuals—lives in a single spreadsheet, able to schedule or export.

Step-by-step guide

Now let’s see AI Sheets in motion. We’ll use open models to unlock the knowledge inside handwritten recipes just like the ones you may find out of your grandma.

Upload your data

Now we have a folder with photos that we are able to simply upload to the app.

The result’s a spreadsheet like this:

Understanding AI actions

Each column in your spreadsheet may be transformed, extracted from, queried, and anything you may imagine using AI actions.

To see this in motion, click on the overlay on top of any column:

Image columns include image operations like extracting text, asking the image, object detection, colorization, adding text, and any custom motion you may consider.

Text columns include summarization, keyword extraction, translation, and custom actions.

A prompt and a model define every AI motion. Let’s see what we are able to do with our handwritten recipes dataset!

Extract text from images.

AI Sheets comes with a template to extract text from images:

The results of this motion is an AI-generated column with the transcribed text. Let’s see an example:

For the above image, the extracted text is as follows:

MEMORANDUM:

From

To

1 Box Duncan Hines Yellow Cake Mix

1 Box fast lemon pudding

2/3 cups water

1/2 cup Mozola oil

4 eggs

Lemon flavoring to taste.

Put in mixing bowl and beat for 10 min.

and REMEMBER... for Quality PRINTING

CALL OR WRITE

Gatling & Pierce

PRINTERS

TELEPHONE 332-2579

22 YEARS OF SERVICE IN NORTHEASTERN CAROLINA

Not bad! But we see it has included printed text for the header and footer, and we’re inquisitive about the recipe text. The explanation this text is included is that we now have used the default template for text extraction, which is as follows:

Extract and transcribe all visible text from the image, including signs, labels, documents, or any written content

Let’s now try a custom prompt.

Here is the extracted recipe details:

– 1 box Duncan Hines Yellow Cake Mix
– 1 box fast lemon pudding
– 2/3 cups water
– 1/2 cup Mazola oil
– 4 eggs
– Lemon flavoring to taste
– Put in mixing bowl and beat for 10 minutes

That is great! But what about more complex images? By default, AI Sheets uses models with a superb balance of speed and accuracy, but you may experiment with hundreds of models. The above example uses the default vision language model Qwen/Qwen2.5-VL-7B-Instruct.

Let’s test a SoTA reasoning model, Qwen/Qwen3-VL-235B-A22B-Reasoning, with a tougher image.

Here’s the comparison between the models:

Each models produce very similar outputs, but with two subtle but essential details (in daring): the temperature and a key ingredient: spinach.

Clean, transform, and enrich text

Once we’re satisfied with the extracted text, we are able to further transform and enrich it. We want to perform an AI motion with the brand new column as follows:

We now have a beautifully structured HTML page for every recipe:

Edit and transform images.

Finally, AI Sheets integrates image-to-image models like Qwen-Image-Edit. This implies you may run AI actions to rework and enrich your images.

For instance, let’s say you need to give your recipes and old-looking style, you’ll want to go to the column and use the B&W template like so:

Result:

Export your dataset

When you’re comfortable along with your recent dataset, export it to the Hub! You’ll be able to export it to a corporation, your personal profile or make it private in the event you don’t need to share it with the community.

You’ll be able to try the dataset we now have just created.

What’s next?

You’ll be able to try AI Sheets without installing or downloading and deploying it locally from the GitHub repo. To run locally and get probably the most out of it, we recommend you subscribe to PRO and get 20x monthly inference usage.

If you may have questions or suggestions, tell us within the Community tab or by opening a problem on GitHub.