Keep a performer’s footage together with Resolve’s IntelliSearch

Three performers share a rehearsal stage. One wears a red coat in the first scene and a dark jacket in the second; another borrows the coat during a break. A search for red clothing alone would mix together two different people. For a short profile of one performer, identity and costume answer different questions.

Blackmagic’s current product page lists individual-face searches alongside object and dialogue searches. The useful task here is to build a reliable collection around one performer, then narrow it to the scene you need.[2]

The rehearsal and performer, Jo, are invented examples. This guide follows the September 2026 Resolve 21.1 manual; we did not analyse footage or observe search results. It assumes your clips are already in the project’s Media Pool.

IntelliSearch requires the Studio edition.[1]

Blackmagic’s US page currently lists Studio 21 at $236, reduced from $295. Check your region’s price before purchasing.[3]

Give the face group a clear reference

Install an IntelliSearch model through DaVinci Resolve > Extras Download Manager. Select it through Media Pool’s three-dot menu > IntelliSearch Mode. In AI Clip Analysis, choose selected clips, enable Analyze for IntelliSearch and Identify faces, then Analyze.[1]

Start with footage from this rehearsal. Decide which person the collection should follow, using a clear shot you can recognise yourself. Call the group Jo because that is the performer’s identity in this exercise, regardless of the character being played or the costume being worn.

Open Media Pool search, then Show Face Gallery > Manage Faces. Name the selected face.[1]

Inspect the group before treating it as an index. In this example, a correct collection should still contain Jo after the costume change, and should exclude the other performer who borrowed the coat. Those two checks test whether the collection follows the person you intended.

Use Improve Faces Grouping Result to confirm matches. Reassign errors by right-clicking and choosing Not [current name], then selecting the correct face.[1]

Do not use costume colour to settle an uncertain identity. Open the actual footage and check the person. If you cannot tell, leave that shot out of your accepted selection until you can establish who it shows. A tidy label is useful only when you trust what it includes.

Add the clue that belongs to the scene

Select Jo’s face, then enter a costume clue. Full Clips shows whole clips; Segments shows matching portions.[1]

For this exercise, try red coat. Treat the query as a proposed input, with no promised result. First ask whether each match shows Jo; then ask whether it belongs to the scene you want. A break between takes may satisfy both person and costume clues while contributing nothing to the performance.

Save a collection you can return to

Add as SmartBin opens the criteria. Name the bin and confirm.[1]

A suggested name is Jo — red-coat scene — rehearsal. Keep a brief note of the grouping correction you made and any shots still uncertain. That note helps distinguish an identification you checked from one you merely inherited from the first analysis.

Report an error or suggest a correction