> For the complete documentation index, see [llms.txt](https://docs.unitlab.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.unitlab.ai/documentation/auto-labeling/batch-auto-labeling-with-prompts.md).

# Batch Auto-Labeling with Prompts

Run prompt-based batch auto-labeling from a workflow, then review results on large geospatial TIFF imagery with tens of thousands of predicted objects.

{% embed url="<https://292810646-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FGjVLUz4wthGkGlRKM6rM%2Fuploads%2FQtSMMa8ATnPtn0ap8HuG%2Fprompt-batch-geospatial-demo.mp4?alt=media&token=eeed31da-bdef-415a-88eb-fd81907a6d91>" %}

[Open the demo in a new tab](https://292810646-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FGjVLUz4wthGkGlRKM6rM%2Fuploads%2FQtSMMa8ATnPtn0ap8HuG%2Fprompt-batch-geospatial-demo.mp4?alt=media\&token=eeed31da-bdef-415a-88eb-fd81907a6d91).

The video walks through the configured prompts, completed queue, and auto-labeled results on large geospatial TIFF imagery. Use it to see the transition from full-image coverage to individual-object inspection.

Describe the objects you want to label, map each prompt to a project class, and let a workflow apply those prompts across your project data. Prompt batch auto-labeling turns text descriptions into editable annotations that move through your normal annotation and review stages.

This guide demonstrates the workflow on **large geospatial TIFF images**, using **house**, **car**, **truck**, and **solar panel** as example prompts. The same setup can target other visual objects that your prompts and project ontology describe.

For prompt-writing basics and interactive detection, see [Prompt Labeling](/documentation/auto-labeling/prompt-labeling.md). For deep zoom and spatial review, see [Geospatial Annotation](/documentation/annotations/geospatial-annotation.md).

## Before you begin

* Add your imagery to a dataset attached to the project, and wait for image preparation to finish.
* Configure the project ontology with compatible **Bounding box**, **Polygon**, or **Mask** classes for the objects you want the model to label.
* Open **Workflows** and prepare a route from **Project** to **Model**, then **Annotate**, **Review**, and **Complete**.

The demo uses segmentation classes for its four object prompts. Choose classes and prompts that match your own annotation goal.

## 1. Configure the Prompt Labelling model stage

Add or select the **Model** stage. Set **Data type** to **Geospatial**, select **All project data** under **Batch**, and choose **Prompt Labelling** in the **Model** field.

Leave **Tile scale** set to **Auto** to begin. Auto uses the image resolution to choose how the large raster is processed. Higher tile-scale values provide a larger field of view for larger objects, with less detail available for small objects.

![](https://292810646-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FGjVLUz4wthGkGlRKM6rM%2Fuploads%2FBjnenCykn2m8nBGmG6K3%2F01.webp?alt=media\&token=3955b8e1-09ea-4d17-b46f-744d45c87db2)

*The demo workflow routes geospatial project data through a Prompt Labelling model stage before human annotation and review.*

## 2. Select classes and enter object prompts

Select each project class you want to auto-label, then enter a short description of what the model should find for that class. In this example, all four classes are selected:

| Project class | Object prompt |
| ------------- | ------------- |
| house         | house         |
| car           | car           |
| truck         | truck         |
| solar panel   | solar panel   |

The **Prompt** field above the class list is optional shared scene context. The demo leaves it empty and uses the individual class prompts. Keep each class prompt focused on one visual concept; use more descriptive wording when a class name alone is ambiguous.

<figure><img src="https://292810646-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FGjVLUz4wthGkGlRKM6rM%2Fuploads%2Fkh0lNI27ymSPQaaWVHjl%2F02.webp?alt=media&amp;token=54832ee4-e405-4693-b872-04d371c9dda5" alt="" width="320"><figcaption></figcaption></figure>

*Four selected classes each have their own detection prompt. The optional scene-context field is separate from these class prompts.*

When the configuration is ready, click **Save and Apply**. The model run starts automatically for eligible images in the selected batch, and the remaining images enter the queue. With **All project data** selected, the workflow processes the matching project images without starting prediction manually on each one.

## 3. Follow the queue and wait for completion

Open **Queues** and use the **Task Queue** to follow items through the **Model** and **Annotate** stages.

For large gigapixel images, AI prediction can take **several minutes**, depending on the image size and the number of objects being predicted. While a large image is being processed, other images may remain **queued**, waiting for their turn as processing capacity becomes available.

During a run, the progress display can show tile and object progress. After prediction, allow the final **Reconciling objects and saving results** stage to finish. Once results are saved, the item moves along the configured route so you can inspect its annotations.

![](https://292810646-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FGjVLUz4wthGkGlRKM6rM%2Fuploads%2F4HbcWvY4Q1kVpyXiWaVK%2F03.webp?alt=media\&token=172d4355-35da-40da-a92a-f80663df2ec3)

*The completed demo queue shows images in Annotate with “Model added” object counts, including 94.5K, 87.4K, and 69.7K objects.*

In this demo, prompt batch auto-labeling predicted **approximately 50–100K objects in some TIFF images**. These are results from the demonstrated imagery; the number of predictions varies with the image content, prompts, and settings.

## 4. Inspect the full-image results

Open a completed item from **Annotate**. Start with the full-image view to check where the model found objects and whether its coverage matches your goal.

Use the object list and class colors to inspect the generated annotations. All annotations shown in this example were produced by the prompt batch run using the configured class prompts.

![](https://292810646-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FGjVLUz4wthGkGlRKM6rM%2Fuploads%2FpWElStyr0agMSsDzAjDg%2F04.webp?alt=media\&token=c2ca2844-25c8-42ff-9489-49f896b9df7e)

*An example large geospatial image after prompt batch auto-labeling. The overview shows automatically generated annotations across the image.*

## 5. Zoom in, correct, and send to review

Zoom into representative areas to check individual houses, vehicles, and solar panels. Inspect boundaries, missed objects, duplicate predictions, and incorrect class assignments, then make any needed corrections.

When the annotations meet your project instructions, click **Send to Review** and continue through the configured workflow.

![](https://292810646-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FGjVLUz4wthGkGlRKM6rM%2Fuploads%2FTMQt4z2SVh3Y0PlomCqp%2F05.webp?alt=media\&token=195406d3-7dec-4595-9412-d015c61bd19d)

*A closer view of the same example result. These object annotations were created automatically by the batch run, and remain editable for human review.*

## Related guides

* [Prompt Labeling](https://docs.unitlab.ai/documentation/auto-labeling/prompt-labeling): write focused text prompts and test interactive detection.
* [Geospatial Annotation](https://docs.unitlab.ai/documentation/annotations/geospatial-annotation): navigate large rasters and review spatial geometry.
* [Batch Auto-Labeling](https://docs.unitlab.ai/documentation/auto-labeling/batch-auto-labeling): understand model stages, queues, and review routes.
* [Batch Auto-Labeling with Find Similar](https://docs.unitlab.ai/documentation/auto-labeling/batch-auto-labeling-with-find-similar): use annotated visual references when examples describe your target more clearly than text.