AutogenAI can read and use images in two distinct ways: as standalone library uploads that get a full AI-generated description, or as visuals embedded inside Word/PDF documents that only get their text extracted via OCR. Understanding this split resolves the most common client confusion around this feature.
1. Standalone image uploads (Library AI)
What it does: Users can upload image files directly to the AutogenAI Library. The Library AI processes the image and generates a full description of its visual content, which then becomes usable throughout the product — Research Assistant, Q&A, and Editor transformations (e.g., using "Expand" to write a description of a chart) can all draw on it.
Confirmed use cases: flowcharts, line graphs, infographics, client testimonials, Gantt charts, and blueprints.
How to use it
- Upload the image file to a library.
- Use Research Assistant to interrogate the image (e.g., ask questions about a chart).
- The source image is shown alongside any AI-generated answer that references it.
- Editor transformations that draw on Library AI can incorporate the image's content.
- "Expand" can be used to generate a written description of a chart or diagram.
- If anonymized on upload, the image itself is not retained — only a redacted text version is stored and shown in the library.
Supported formats
gif, jpg, jpeg, png, webp. CAD/DWG files are not supported. Maximum file size follows the standard 2GB library upload limit.
Upload method
Standalone image files must be uploaded directly to the library. They are not processed via SharePoint sync, to avoid overloading the document-processing pipeline with large image batches.
Regional availability
Live for all regions, including Australia. AU was initially blocked because Azure did not offer GPT-4o in-region; this was resolved and the feature is confirmed fully live for AU as of mid-2026. Note: very detailed drawings may render slightly less precisely depending on the model used.
2. Images embedded inside Word/PDF documents (OCR only)
An image buried inside a Word document or PDF is not described the same way a standalone upload is.
- Embedded images go through OCR: any text sitting on the image is extracted and made searchable — but the platform does not describe or interpret what the visual actually shows.
- This OCR step works even when the parent document is synced in via SharePoint (unlike standalone image uploads).
- Recommendation: if the visual meaning matters (e.g., a chart, blueprint, or annotated site plan), it should be uploaded as its own standalone image file rather than left embedded in a document.
3. Scanned / image-based PDFs
AutogenAI has OCR capability for scanned PDFs (where the page is stored as an image rather than selectable text). In practice, some scanned PDFs still fail with ERROR_NO_TEXT_FOUND because full-document OCR was scaled back. Recommended workarounds when this happens:
- Re-export or regenerate the source as a text-based PDF (from Word, Google Docs, etc.).
- Run it through external OCR software (Adobe Acrobat Pro, Microsoft Office Lens, or a free OCR tool), then re-upload.
- Upload as a Word (.docx) file if available — these extract reliably.
Limitations summary
| Limitation | Detail |
|---|---|
| No image generation | AutogenAI reads and extracts from images; it does not generate images or convert text into infographics. |
| SharePoint sync | Standalone image files must be uploaded directly to the library — this is not supported via SharePoint sync. |
| File formats | gif, jpg, jpeg, png, webp only. CAD/DWG and other proprietary formats are not supported. |
| GIS data | Raw GIS formats (.shp, .kml, .tif) are not supported. GIS data exported as PNG/JPG/PDF maps can be processed. |
| Scanned PDFs | Full-document OCR can fail (ERROR_NO_TEXT_FOUND) on image-only PDFs — see workarounds above. |
FAQ
Which image formats are supported?
gif, jpg, jpeg, png, webp.
Will embedded images in Word/PDF get an AI description like standalone uploads?
No — only OCR'd text, not visual description.
Is this available in Australia?
Yes, fully live as of mid-2026.
Can I upload CAD/DWG or GIS files directly?
No — not currently supported.