The short answer

ENCORE runs semantic search over your clip library using prompt embeddings + Brand Brain signal. Top-K matching clips appear with violet selection borders. You can accept, override, or hand-pick before rendering.

What ENCORE matches on

  1. Prompt semantics. "Morning routine" matches kitchen, coffee, breakfast clips.
  2. Recency. Newer clips get a small boost. Most users prefer fresh content.
  3. Prior performance. Clips that produced high-engagement renders before get a boost.
  4. Brand Brain pacing. Clips that fit your pacing rhythm get a boost (longer clips for slow pacing, shorter for fast).
  5. Visual diversity. ENCORE avoids stacking similar shots back to back.

Manual override

In Library Mode, multi-select up to 12 clips manually. They appear with violet borders. Tap "Use N clips" at the bottom. ENCORE uses your selection instead of its auto-pick.

Tagging clips for better matches

Optional but powerful: tag your clips with semantic labels (e.g., #workout, #podcast, #tutorial) inside the source app or via the Library Mode tag panel. Brand Brain weights tagged clips higher when prompts reference matching topics.

What to do when the wrong clips get picked

  1. Check the prompt: was it specific enough? "Make a reel" is too vague. "60-second reel about cold-weather running" is sharp.
  2. Check Brand Brain depth on the Audience Signal: if low, the system has less data to bias selection. Train more.
  3. Manual override. Pick the right 3-5 clips and ENCORE uses those.

Related