Cutsio Blog

Cloud Dailies for MOS Footage: Why Transcript Search Fails on Cinema RAW

Why transcript-based search cannot handle MOS (motor-only sync) cinema raw footage, and how visual search and Visual Intelligence solve the problem for B-roll, inserts, plates, and silent footage.

Why does transcript search fail on MOS cinema raw footage?

Transcript-based search requires audio content to index. MOS (motor-only sync) footage — action shots, B-roll, inserts, visual effects plates, establishing shots, and any clip without usable scratch audio — has no spoken content to transcribe. Cinema raw formats like ARRIRAW and RED R3D are frequently shot MOS, especially for high-frame-rate slow-motion, gimbal work, and aerial footage. Visual Intelligence helps by indexing visual content independently of speech, making MOS footage searchable by visible objects, scenes, actions, and composition cues.

The limitation of transcript-based search in platforms like Frame.io and Dropbox Replay is fundamental: they analyze the audio track and create a text index of spoken words. If there is no audio — or if the audio is unusable scratch track from a wind-blown microphone or a camera's internal mic in a noisy environment — the search index is empty. The footage is effectively invisible to search.

For cinema raw productions, MOS footage represents a significant portion of the total capture. Action sequences, slow-motion shots, second-unit photography, aerial footage, underwater footage, and visual effects plates are all typically shot MOS. On a feature film, 30-50% of the total footage may have no usable audio. A search system that cannot handle MOS footage leaves half the footage unsearchable.

Search your video library faster with How to Search Your Entire Video Library by Meaning.

How does Visual Intelligence index MOS footage differently?

Cutsio's Visual Intelligence analyzes visual content independently of audio. It uses computer vision models to identify searchable cues such as objects (car, person, building), scenes (forest, interior, city street), actions (running, climbing, driving), compositions (close-up, wide shot, aerial), and visual characteristics (warm lighting, blue sky, night scene).

The index is created from the review stream, which is generated from the original camera files. For MOS footage, the visual index is the complete search record — there is no audio transcript to supplement it.

| Search Feature | Transcript-Based | Visual Intelligence |

| :--- | :--- | :--- |

| Searches MOS footage | No (no audio to transcribe) | Yes (analyzes visual content) |

| Finds objects | No | Yes (car, person, building, etc.) |

| Finds scenes | No | Yes (forest, interior, city street, etc.) |

| Finds actions | No | Yes (running, driving, climbing, etc.) |

| Finds compositions | No | Yes (close-up, wide shot, aerial, etc.) |

| Finds spoken content | Yes | Yes (via transcript) |

| Finds visual characteristics | No | Yes (warm lighting, day/night, color) |

The practical difference: a search for "find the wide shot of the car driving through the forest at sunset" returns results from MOS footage because Visual Intelligence can identify the car, the forest setting, the wide composition, and the warm sunset lighting — all from the visual content alone.

What types of cinema raw footage are most affected by the MOS search gap?

Different production types have different proportions of MOS footage. Understanding where the gap hurts most helps prioritize Visual Intelligence in the dailies workflow.

| Production Type | Typical MOS Percentage | Most Affected Footage Types |

| :--- | :--- | :--- |

| Narrative feature | 30-50% | Action scenes, B-roll, establishing shots, VFX plates |

| Commercial | 40-60% | Product shots, slow-motion, beauty shots, aerials |

| Documentary | 10-20% | B-roll, timelapse, atmospheric footage |

| Music video | 50-70% | Performance MOS, concept shots, slow-motion |

| Sports broadcast | 40-50% | Slow-motion replays, crowd shots, wide coverage |

For narrative features and commercials — the primary users of cinema raw formats — the MOS percentage is high enough that transcript-only search is severely limited. A DIT or assistant editor searching for a specific take of a car crash in an action sequence cannot use transcript search because the crash was shot MOS for sound design flexibility.

playback-id="IRBqKFllfQTZRgUpvF00DnjqMROLtyclqpWYRLQez6KQ"

title="Cutsio Visual Intelligence — search video by what the camera saw"

poster="https://image.mux.com/IRBqKFllfQTZRgUpvF00DnjqMROLtyclqpWYRLQez6KQ/thumbnail.jpg">

How does Agentic Chat handle MOS queries?

Agentic Chat in Cutsio understands natural language queries about visual content and searches the Visual Intelligence index regardless of whether the footage has audio.

Examples of MOS queries that Agentic Chat can handle:

• "Show me the wide shot of the building exterior from Day 4"
• "Find all the close-up product shots with blue background"
• "Which takes have the actor jumping over the fence?"
• "Find the aerial sunset footage from the second unit"
• "Show me the VFX plates of the city skyline"

The queries work because they describe visual content — objects, scenes, actions, and composition — not spoken words. Agentic Chat searches the visual index created by Visual Intelligence, which exists independently of any transcript.

For clips that do have audio, Agentic Chat searches both the visual index and the transcript simultaneously, returning results from either or both sources.

How does the MOS search gap affect the dailies review workflow?

The MOS search gap creates specific workflow inefficiencies in traditional dailies pipelines:

• Manual scrubbing: Without searchable MOS footage, editors and assistants must manually scrub through hours of B-roll and action footage to find specific shots
• Missed selects: Good takes in MOS footage can be overlooked because there is no efficient way to find them
• Duplicate effort: Multiple team members may scrub the same MOS footage independently, each looking for different shots
• Delayed conform: When the picture is locked, finding the exact MOS clips for conform requires matching timecode notes from the offline edit — a fragile process

Visual Intelligence reduces these inefficiencies by making MOS footage searchable from the moment it is ingested. The assistant editor can type "find the car crash wide shot" and review visual matches from the MOS action footage instead of scrubbing blindly.

How do you set up Visual Intelligence for MOS-heavy productions?

The setup is identical for MOS and audio-track footage. Visual Intelligence analyzes the generated review assets from clips uploaded through Cutsio's enterprise raw ingestion add-on, regardless of audio content.

1. Upload native ARRIRAW or RED R3D files through the enterprise add-on
2. Cutsio generates streamable review assets from the original camera files
3. Visual Intelligence indexes visual content from the review assets
4. The index includes objects, scenes, actions, and visual characteristics
5. Agentic Chat searches the visual index for natural language queries
6. MOS and audio-track footage are searchable through the same interface

No special configuration is needed for MOS footage. Visual Intelligence indexes the visual content automatically. The only difference is that MOS clips will not have a transcript index, so visual search terms and clip notes matter more.

FAQ

Does Visual Intelligence work on footage that has audio but no dialogue?

Yes. Even when footage has audio — wind noise, ambient sound, music — Visual Intelligence indexes the visual content independently. The visual index is not dependent on audio quality or content.

How accurate is Visual Intelligence on MOS action footage?

Visual Intelligence is most accurate on clearly visible objects, scenes, and actions. Fast motion, extreme close-ups, and heavily obscured subjects may produce less precise results. Accuracy improves with higher resolution source footage and well-lit scenes.

Can I search for specific camera moves in MOS footage?

Visual Intelligence can help surface shots with visible motion and composition cues, but camera-move search should be treated as an assistive discovery tool rather than a substitute for editorial review. For precise movement categories such as pan, tilt, or tracking, verify results by watching the returned clips.

Does Cutsio search MOS footage differently from footage with audio?

No. The search index is created from visual content regardless of audio. Footage with dialogue can also have a transcript index, but MOS clips still remain searchable through visual cues.

What happens if I search for dialogue in a MOS clip?

Agentic Chat will return no transcript matches for the MOS clip, but it will return visual matches. For example, searching "find the scene where the actor speaks" on MOS footage would return no transcript matches but might return visual matches if the actor's face and mouth movement are visible.