Cutsio Blog

How to Search RED RAW Footage by What's in the Frame

Stop scrubbing through RED R3D clips to find specific frames. Learn how Visual Search lets you search RED RAW footage by describing objects, actions, scenes, and lighting — instantly.

How do you search RED RAW footage by what's in the frame?

Upload RED R3D originals to Cutsio, which generates streamable review assets, indexes every frame with Visual Intelligence, and lets you search the entire library by describing the visual content — objects, people, scenes, actions, and lighting — in natural language. Find a "sunset silhouette shot of cowboy on horseback" or "close-up of hands opening a briefcase" across terabytes of R3D footage in seconds.

RED RAW is the proprietary raw format from RED Digital Cinema cameras — V-RAPTOR, KOMODO, KOMODO-X, and DSMC2 — stored in the .r3d container. A single narrative feature shot on RED can generate 40 to 80 TB of R3D files across 30 to 50 shoot days. Finding a specific frame in that library by scrubbing through clips in a traditional NLE is like finding a needle in a haystack while wearing a blindfold.

Traditional search methods depend on DIT camera reports, handwritten scene notes, and folder hierarchies. All three are unreliable. Camera reports only describe what the script supervisor intended, not what the camera actually captured. Scene notes get lost or misinterpreted. Folder hierarchies break down when different editors organize the same footage differently.

Cutsio's Visual Search solves this by analyzing the actual visual content of every frame. Through the enterprise R3D ingestion add-on, you upload the native .r3d files, Cutsio transcodes them into streamable review assets, indexes every frame with computer vision, and retains the original R3D files as attachments for conform. The result is a RED RAW library you can search by what the camera literally saw.

Working with raw camera footage? Check out How to Build a Searchable Library From ARRIRAW, RED R3D, and ProRes Footage.

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

Why is searching RED RAW footage by visual content better than using camera reports?

Searching RED RAW footage by visual content is better than using camera reports because camera reports describe the director's intent, not the actual footage — and intent does not match reality when takes go wrong, blocking changes, or lighting shifts unexpectedly.

Camera reports are written by the script supervisor or DIT during the take. They describe what was supposed to happen: "Scene 24, Take 3, Master Shot, Dialogue." But they do not describe what actually ended up in the frame. Consider these common scenarios:

| Scenario | Camera Report Says | What Actually Happened |

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

| Blocking change mid-scene | "Master shot" | Actor crosses to the wrong mark, blocking changes |

| Lens flare or artifact | "Take 5 — good" | Flare obscures the actor's face for three seconds |

| MOS action sequence | "Scene 42 — car stunt" | No context on whether the car enters frame left or right |

| Lighting shift | "Golden hour" | Clouds rolled in, lighting is flat and overcast |

Visual Search bypasses the unreliable middle layer of human note-taking. It analyzes the pixels themselves. When you search for "actor crosses in front of window with lens flare," the system returns the actual frames matching that visual description — regardless of what the camera report says about that take.

How does Visual Search work with RED R3D footage specifically?

Visual Search indexes every frame of the RED R3D review stream using computer vision models that recognize objects, scenes, actions, faces, spatial relationships, and lighting conditions — then lets you query the entire R3D library using natural language descriptions that match what a director or editor would say.

The indexing pipeline works as follows:

1. Frame-by-Frame Analysis: During ingestion, Cutsio's Visual Intelligence engine analyzes every frame of the R3D review stream. For RED footage, this analysis works even on clips that have no scratch audio — a common scenario for RED productions shooting MOS.
2. Concept Detection: The system detects objects (cars, horses, weapons, furniture, props), scenes (desert, office, nightclub, kitchen, forest), actions (walking, running, fighting, driving, kissing), and composition (close-up, medium shot, wide shot, over-the-shoulder).
3. Semantic Vector Indexing: Each frame is embedded into a vector space that maps visual concepts semantically. A search for "gun on table" will also return frames where a pistol or rifle appears on a desk, even if "gun" was not the specific label assigned during indexing.
4. Natural Language Query Matching: When you type a search, the system compares your query against the frame vectors and returns the most visually similar results, ranked by relevance.

What specific visual queries work best with RED R3D footage?

The most effective queries for RED RAw combine multiple visual elements. Examples that return precise results:

• "Wide shot of desert landscape with jeep driving toward camera"
• "Close-up of actor's hands loading a revolver"
• "Night exterior with rain and neon reflections"
• "Over-the-shoulder shot of two actors arguing at a dining table"
• "Low-angle shot of building exterior establishing"
• "Slow-motion shot of glass breaking — 120 fps or higher"
• "Dolly shot following actor walking down hallway"

Each of these queries describes visual content that a camera report would never capture in enough detail. The system understands these as visual concepts, not keyword tags.

playback-id="IRBqKFllfQTZRgUpvF00DnjqMROLtyclqpWYRLQez6KQ"

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

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

How do you search RED RAW footage by technical metadata combined with visual content?

Beyond visual search, Cutsio indexes technical metadata from the R3D headers — including frame rate, resolution, ISO, color space, and compression ratio — so you can combine visual queries with camera-specific filters for precision search.

RED R3D files contain rich technical metadata in their headers. Cutsio preserves and indexes this metadata alongside the visual content, enabling queries like:

• "Slow-motion shots of car crash at 120 fps"
• "Close-ups shot at 800 ISO or higher — low-light scenes"
• "Wide masters in REDWideGamutRGB color space"
• "Clips from KOMODO at 6K resolution"
• "Takes with REDCODE compression ratio of 8:1 or lower"

Combining visual and technical queries creates powerful precision: "Find the 120 fps slow-motion shots of the stunt double performing the fall, from the KOMODO-X." This single query uses Visual Search (understanding the visual action), metadata search (filtering by frame rate and camera model), and organizational context (knowing which shoot day contains the stunt work).

How do Collections improve RED RAW search workflows for editorial teams?

Collections in Cutsio let you save search results into curated visual hubs that the entire editorial team can access, search, and share — so every time you discover a great frame through Visual Search, you can immediately add it to a selects reel without exporting or renaming files.

The workflow for editorial teams is:

1. The assistant editor uses Visual Search to find all matching frames for a specific scene — "two-shot argument scene from Day 12."
2. The assistant editor creates a Collection called "Scene 24 — Selects" and adds the matching clips.
3. The editor opens the Collection, reviews the selects, and marks preferred takes.
4. The director reviews the Collection through a share link and leaves timecoded comments.
5. The assistant editor exports a selects EDL from the Collection, referencing the original R3D file names and timecodes.

This replaces the traditional workflow of exporting individual clips, renaming them, organizing them into folders, and sending them via email or shared drive. Everything stays in Cutsio — searchable, shareable, and linked back to the original R3D files.

How does Agentic Chat make RED RAW search accessible to non-technical team members?

Agentic Chat in Cutsio is a conversational AI interface that lets any team member — director, producer, editor, or client — search RED RAW footage using plain English questions without learning search syntax or navigating folder structures.

The production team simply asks:

• "Show me all the dolly shots from Day 7."
• "Which takes from Scene 42 have the actor walking through the door?"
• "Are there any close-ups of the briefcase being opened?"
• "Find the shot where the lighting changes from warm to cool mid-scene."

Agentic Chat combines Visual Search, metadata search, and Collection context to return precise results. For RED footage specifically, it can also query technical R3D header metadata — frame rate, compression ratio, ISO, white balance, and color space — and combine those with visual queries. "Find the 6K KOMODO-X clips shot at 48 fps where the actor runs through the frame."

How do you maintain the RED color pipeline while searching by visual content?

Cutsio generates the review stream with REDWideGamutRGB and Log3G10 color science applied during the cloud transcode, while the original R3D files remain untouched for the final grade — so the visual search experience accurately reflects what the colorist will work with.

When the DIT sets a look on set using RED's in-camera tools or software like Silverstack, that look is stored as metadata in the R3D file. Cutsio applies this look metadata during the review asset generation where possible. The search thumbnails and playback show the footage with the intended color treatment, making it easier to evaluate shots visually while searching.

When the editorial team exports selects for the conform, they work from the original R3D files — completely unmodified. The Visual Search and review process never touches the original sensor data.

How does Storage pricing work for searchable RED RAW libraries?

Cutsio's Storage uses a pay-for-minutes model that separates storage cost from file size — critical for RED RAW libraries that routinely exceed 40 TB per production.

Traditional cloud storage charges per gigabyte. A 40 TB RED production stored on Google Drive or Dropbox can cost $800 to $1,600 per month. With Cutsio, the same production pays based on the total minutes of footage, making large RAW libraries significantly more affordable to keep online and searchable.

The review assets remain streamable and fully searchable through Visual Intelligence at all times. The original R3D files are retained as attachments for download and conform when needed.

FAQ

Can Visual Search find frames in RED footage that has no audio?

Yes. Visual Search indexes visual content only, so MOS footage is fully searchable. This is one of its primary advantages over transcript-based search tools that cannot handle clips without dialogue.

Does Visual Search work with REDCODE RAW at all compression ratios?

Yes. Visual Search indexes the review stream, which is generated from the R3D file regardless of the REDCODE compression ratio used during recording. The search accuracy is independent of the compression setting.

How do I get access to RED R3D ingestion for my production?

RED R3D ingestion is available as an enterprise add-on for qualified production accounts. Contact the Cutsio sales team to enable raw format support and Visual Search indexing for your RED production.

Can I search across multiple RED productions simultaneously?

Yes. If you have multiple RED projects in your Cutsio library, you can search across all of them at once using Visual Search, or restrict your search to specific Collections or projects.

What happens to my RED RAW search index after the production wraps?

Your footage and its search index remain in your Cutsio library for as long as your account is active. The original R3D files and review assets are retained. If you need to revisit the project for additional edits, the search index is ready to use immediately.