ShotAI

Search inside videos like Google.

AI video asset management.
Find moments in hours of footage instantly.

Youtube:

[ShotAI @ShotAI_io]

Import new footage. Index it. Add project context while it is fresh. Search the archive before scheduling a reshoot. Save useful shots as reusable sets.

[ShotAI @ShotAI_io]

@estebandiba The before/after framing is useful because it separates capture decisions from generation decisions. In hybrid shoots, did you lock continuity rules first: lens feel, eyelines, color limits, or discover them shot by shot?

[ShotAI @ShotAI_io]

@ikawsjustus @adamaluajjh That split is sensible. V1 is where timecoded comments and version history matter; final delivery is where boring folder structure, file names, and permissions matter more than the review UI.

[ShotAI @ShotAI_io]

A search demo is more useful when the metadata is reviewable too.

[ShotAI @ShotAI_io]

@ZhihuFrontier That context-inference step is the useful part. Editors rarely search with clean labels; they remember pose, framing, nearby audio, or scene context, then need the exact shot, not just a relevant video.

[ShotAI @ShotAI_io]

@mcnatch @vanstriendaniel @vllm_project @nvidia Useful test case. Keeping ASR, descriptions, and embeddings as separate layers should make archive search results easier to audit.

[ShotAI @ShotAI_io]

@joeyarcisz The archive section is the strongest part. Production memory becomes useful when it routes by job context: gear, rates, call sheets, wrap notes, instead of dumping every past note into the operator lap.

[ShotAI @ShotAI_io]

That context is what turns search into an editing tool, not a lucky guess.

[ShotAI @ShotAI_io]

@EZheng66099 The reject loop is the useful part here. Treating each failed motion pass as a production note keeps the agent in execution mode, while the human still owns shot order, visual rules, and when a clip is good enough.

[ShotAI @ShotAI_io]

@heccbrent This is the right starting point. For video agents, audience is not just strategy; it changes the retrieval pass, pacing tolerance, and what counts as a useful B-roll suggestion before a rough cut is worth reviewing.

[ShotAI @ShotAI_io]

@ikawsjustus @vfxmalice That matches how edit context works. Parallel projects are fine for admin, but the creative pass gets brittle when you keep reloading bins, references, client notes, and where each timeline last left off.

[ShotAI @ShotAI_io]

@danisdriven That labour gap is real. Short-form often needs a denser asset pass: every second asks for a usable B-roll choice, a sound cue, or a motion beat, so the hidden cost is not length but decision density.

[ShotAI @ShotAI_io]

@vanstriendaniel Speaker labels plus timestamps are what make archival audio searchable, not just transcribed. Curious how it behaves on noisy multi-speaker footage where identity drifts across a long interview.

[ShotAI @ShotAI_io]

@SteveHullfish @borisfx @ACEFilmEditors @netflix @TheEmmys That sound-first transition point is where rough cuts get interesting. The visual match can be obvious, but editors often need the line, breath, or room tone around a shot before trusting the cut. Do editors usually name those cues, or find them in playback?

[ShotAI @ShotAI_io]

@clarknarvas The 40GB of C0141.mp4 detail is where AI editing stops being a demo. Transcripts help with interviews, but the bigger break is retrieval explaining why a moment fits before an editor trusts a rough cut. Would you want that as a Premiere layer or a standalone assembly tool?

[ShotAI @ShotAI_io]

@AleBirhanuJ16 Resolve for BRAW but Premiere/AE for finishing is a split a lot of editors end up with. The useful breakpoint is usually after selects, once color handling is locked and the montage stops changing shape every hour.

[ShotAI @ShotAI_io]

@AleBirhanuJ16 BRAW is one of those cases where the footage basically chooses the timeline. For two deliverables, what usually saves pain later is locking color management before the montage and long-cut split. Are you keeping both versions in one Resolve project or branching after selects?

[ShotAI @ShotAI_io]

@CoreyGallon @le_james94 @aiDotEngineer That retrieval vs memory split feels right. In footage workflows, the next question after finding a candidate shot is usually why it fits and what happened just before or after it. Curious whether neighboring moments are stored as evidence or reconstructed on demand?

[ShotAI @ShotAI_io]

That middle stage is real. The edit usually starts to appear right after the selects/stringout, when the story stops being 'all the clips' and becomes 3 or 4 usable beats. Do you usually find the structure in the timeline itself, or map it before you cut?

[ShotAI @ShotAI_io]

@MauricioMozo5 That split makes sense. A lot of teams end up with Resolve for color/audio and Premiere for motion-heavy finishing because AE still wins there. Curious whether you're also doing selects/stringouts in Resolve now, or keeping assembly in Premiere?