Media scansCopy Anchor Link

The Scan as dropdown next to every scan action lets you choose what Claira sends to the model:

Auto is selected by default. For each document, Claira looks at the actual files attached in Nuix Discover. If any of them is a PDF, image, audio, or video file that the model can read directly, Claira sends that file. Only when no such file exists (common for Word documents, Excel sheets, emails, and HTML) does Claira fall back to the extracted text.

Switch to one of the explicit modes when you want Claira to use the same source across every document in a run, regardless of what's on each one.

How Auto picks per documentCopy Anchor Link

For each document, Auto applies one rule:

1. If the document has a PDF, image, audio, or video file attached, send that file. Claira walks the document's content files in priority order and sends the highest-priority one the model can read natively. PDFs always go to Image scan: the PDF itself is sent to the model, whether it was a digital-native export or a scanned image. Native images go to Image, audio files to Audio, video files to Video.
2. Otherwise, send the extracted text. Word documents, Excel sheets, emails, HTML pages, plain .txt files, and other formats the model can't ingest directly are reviewed against the text Nuix extracted from them.

That's it. There is no separate "scanned vs. native PDF" detection anymore. If a PDF exists on the document, Claira sends the PDF. This means the model sees layout, signatures, redactions, handwriting, and any other visual content the OCR layer might miss.

Text fallback when the file cannot be used. Auto chooses a document's file from the file list Nuix Discover shows, so some problems only come to light when Claira opens the file: it is over the file size limit for the scan, its contents are not really the format its extension promised, or the model you picked does not read that format. In Auto, none of those fail the document. If the document has extracted text, Claira reads the text instead, charges it at the Text rate, and records Text as the mode it used. When the document has no extracted text either, Claira marks it as skipped and carries on with the rest of the run: you'll see the reason on the document in the task history.

One problem has no text fallback. A file can pass all of those checks and still be one the AI model cannot read once it receives it: a PDF with no pages, for example, or a damaged or password-protected file. In a scan that runs in your browser, Claira does not read the text instead, even when the document has some. A bulk scan marks the document skipped with an E-FETCH note and does not charge it, and a single scan stops with the message. To scan the document's extracted text, run it again with Scan as set to Text.

This fallback belongs to Auto alone. In an explicit Image, Audio or Video run you asked for the file, so a document with no file that mode can read is marked skipped rather than quietly read as text. Skipped documents are refunded either way.

Bulk task history shows the Auto badge on the run, and each document records the mode Auto actually picked so you can audit the mix after a run completes.

When to pick an explicit modeCopy Anchor Link

Image scanCopy Anchor Link

• Scanned PDFs with little or no extracted text
• Handwritten notes and notepads, including scans stored as TIFF or JPEG 2000 images
• Photo-based records or screenshots
• Documents where layout and visual structure matter (forms, redactions, signatures)

With a Gemini model, Image scan accepts: PDF (.pdf), PNG (.png), JPEG (.jpeg / .jpg), WebP (.webp), GIF (.gif), BMP (.bmp), HEIC (.heic), HEIF (.heif), TIFF (.tif / .tiff), JPEG 2000 (.jp2).

Claira converts TIFF and JPEG 2000 images to a readable format before the scan, and a multi-page TIFF is scanned as one document of up to 500 pages. Nuix Discover records a file's real format when it is ingested, so a scan saved with a .pdf name can still be a TIFF or JPEG 2000 image in the case. Its Document Type field shows which. PDFs that carry extra data before the PDF header, which some scanners and scanning apps add, are read as normal PDFs.

Approximate max length per document: ~900 pages (for PDFs).

Audio scanCopy Anchor Link

• Voicemail and call recordings
• Recorded interviews or dictation
• Any document whose primary content is spoken audio

Audio scan accepts: MP3 (.mp3), WAV (.wav), AAC (.aac), OGG (.ogg), FLAC (.flac), M4A (.m4a), AIFF (.aiff).

Approximate max length per document: ~8 hours of audio.

Video scanCopy Anchor Link

• Surveillance clips, body-cam recordings
• Recorded depositions or hearings
• Video evidence with both visual and audio relevance

Video scan accepts: MP4 (.mp4), MOV (.mov), WebM (.webm), AVI (.avi), MKV (.mkv), MPEG (.mpeg), 3GP (.3gp), FLV (.flv), WMV (.wmv).

Approximate max length per document: ~40 minutes of video (with audio track).

Which models can run a media scanCopy Anchor Link

Every Gemini model runs all three media modes, with every format listed above. Three other models run Image scans on a narrower set of files, and the rest review document text only.

With Llama 4 Maverick, Gemma 4 31B or Qwen 3.8 Max selected:

• Auto sends a document's image when the model can read it, and its extracted text otherwise. A PDF, for example, is reviewed from its text.
• If you choose Image and a document's file is a PDF, a multi-page scan or another format the model does not read, that document is not scanned, and the message names the file type. Run it again as Text, or switch to a Gemini model.
• Audio and Video need a Gemini model.

If the selected model cannot run the mode you chose at all, Claira stops before anything is sent to the model and offers a Change model button that opens the model picker.

How to run a media scanCopy Anchor Link

1. Open Single Review, Bulk Scan, or Multi-Code.
2. Leave Scan as on Auto to let Claira pick per document, or switch to a specific mode for the whole run.
3. Run the scan.

File size limitsCopy Anchor Link

How large a file a scan can read depends on where the scan runs:

When a document's file is over the limit, Claira tells you the file's size and the limit, sends nothing to the AI and charges nothing. What happens next depends on Scan as:

• Auto reads the document's extracted text instead, at the Text rate. If the document has no extracted text, a single scan stops with the message and a bulk scan skips the document.
• Image, Audio and Video do not fall back to text. A single scan stops with the message, and a bulk scan skips the document with an E-FETCH note.

To scan a file between 23 MB and 50 MB, select the document in Nuix Discover, start a bulk scan and check Run as background processing task. To scan the document's text instead, switch Scan as to Text. See Scan says the document's file is too large.

Running media scans in the backgroundCopy Anchor Link

On a case where Background Processing is set up, a bulk Image, Audio, Video or Auto scan can run as a background processing task, single-field or Multi-Code, just like a Text scan. Check Run as background processing task in the bulk scan confirmation dialog, or click Send to background on a scan already running in your browser.

• Claira uploads the files first. Each document's file is uploaded from your browser before the scan runs on the servers. Keep Claira open until the bulk scan card's Uploading files for Task #NNN progress finishes; after that you can close it. If the upload is interrupted, anyone working in the case can choose Finish uploading to complete it. See Uploading files for image, audio, video and Auto scans.
• A higher file size limit and the same price. A background scan reads files up to 50 MB, against 23 MB for a scan in your browser (see File size limits), and TIFF and JPEG 2000 images are converted automatically. A background media scan costs the same tokens as the same scan in your browser, and Auto charges each document at the type it resolved to. A document Claira could not read gets an E-FETCH note and is not charged either way, but the two kinds are reported differently: a document with no file the scan can read, or a file over 50 MB, is skipped (in Auto, only when it has no text to fall back on), while a file that could not be downloaded, could not be converted, or was never uploaded is failed.
• Agent Mode scans stay in your browser. An Image, Audio, Video or Auto scan the Agent starts always runs in your browser; only text scans started by the Agent can run in the background.

When Scan as is stuck on TextCopy Anchor Link

On a Claira Private case, whether it runs on the shared connection or on your local model, Claira sends text and nothing else to the model: Scan as shows Text and cannot be changed, and a short note beside the control says why. Image, audio and video scans are not available in Claira Private.

Your saved choice is never overwritten: Claira keeps showing Text only on that case. The note appears anywhere you pick a scan type, in Prompt Lab, Bulk Scanning and Multi-Code.

Smart error suggestionsCopy Anchor Link

If a Text scan fails because the document has no extracted text, but Claira sees an image / audio / video file on the document, the error message in the field will say "this looks like audio content. Try Audio scan" (and similarly for image/video). Switch the dropdown and re-run. When you're on Auto, Claira makes this switch for you on the fly: the error suggestion only appears when you've pinned the dropdown to Text explicitly.

Plan availabilityCopy Anchor Link

Every scan mode is available on every plan, including Starter, single document and bulk. Auto, Text, Image, Audio, and Video can all be used by every customer without an upgrade prompt. Single scans are free in every mode. In a bulk scan, the per-document token cost (1, 5, 10, or 20 tokens) still applies and is deducted from your case's token balance as the scan runs; how many tokens your plan grants each month is what limits volume, not the mode.

When you pick Auto for a bulk run, Claira reserves the worst-case rate (20 tokens / doc) up front and refunds the difference as each document resolves to a cheaper mode. You pay only for what each document actually used, unless your plan renews before the task ends: see What a bulk scan refunds.

---

Need help? Contact us at support@claira.to.