Back to Playlist · VidFarm Walkthrough Tutorial
Part 20 of 33Edit Demo — Easy Remix7 min read
Edit demo #3 — easy remix with raws

Remix a template with clips you hunt

Point the AI at a long podcast or YouTube video, let it hunt the best short clips, then drop those raws into a remix template and cut them to the beat — all by chatting.

Copies the whole guide — prompts, next steps & all — for your AI agent
Watch this chapter
The remix demo: hunt short clips out of a long-form video with an AI scan, watch it run in the cloud-status panel, then swap the hunted raws into a template's scenes and render.

Key takeaways

  • An AI hunt (/clips/scan) mines a long source for the best short, on-topic clips — async, in the cloud.
  • The hunted raws land in your Library → Raws; you drop them into a template with set_layer_media.
  • Need a surgical single cut instead of a hunt? /raws/clip-range grabs one exact in/out span.
  • The editor auto-forks and the copilot confirms before rendering — same safe loop as every demo.

The idea: mine long video for short gold

A remix is a scene replace: you keep the template's rhythm, transitions, and caption cadence, but you fuel it with fresh footage cut from a long source — a podcast, a VOD, a YouTube upload. Rather than scrub an hour of video by hand, you tell the AI what you're looking for and it hunts the clips for you.

A long grey film reel stretches out to the left representing a long podcast or VOD; a pair of scissors snips segments out of it and those segments turn into glowing golden vertical phone-video nuggets that fly into a little basket on the right, glittering with honey-gold light.
Cut long video for short gold. Instead of scrubbing an hour of footage by hand, an AI hunt mines a long source for the few short, on-topic clips worth keeping.

Two routes do the sourcing. /clips/scan is the AI hunt — it watches a long-form source and returns several short clips that match your brief (async; it bills only AWS compute, and the AI tagging runs on your own keys). /raws/clip-range is the surgical cut — one exact in/out span, no AI. This demo leans on the hunt. For the full sourcing playbook — the public raws feed, importing by URL or upload, and the clipper — see Sourcing & clipping raws with AI .

Manual walkthroughDo it yourself (the web UI)

  1. Fork a remix-style template

    On the Discover feed , open a fast-cut remix format and click into the Trackpad Editor . It auto-forks on your first edit.

    Open Discover feed

  2. Ask the AI to hunt clips from a source

    Paste a long-form URL and describe what you want cut out of it:

    "Hunt 9:16 clips about cold plunges from this video: <YouTube/podcast URL>. I want about 5 clips around 3 seconds each, prefer scenes without on-screen text."

    The copilot POSTs /clips/scan with your source, prompt, aspect, a soft target duration, and "avoid text." It returns immediately with a scan id and runs in the background.

  3. Watch the hunt run

    Hunts are async — don't wait on the chat. Open the cloud-status panel (#pop-panel/cloud-status) to watch the scan progress. When it finishes, the clips show up as raws.

  4. Review the hunted raws

    Browse your new clips in Library → Raws (they also surface in the public raws feed view). Pick the ones you like.

    Open Library → Raws

  5. Swap the raws into the template's scenes

    Tell the copilot to slot them in, keeping the template's timing:

    "Swap scenes 1–4 for the hunted clips, keeping each scene's timing and the transitions."

    That's set_layer_media per scene using each raw's view_url — or add_layer / generate_layer intent:add for a net-new scene.

  6. Recaption, review, render, approve

    Recaption to your angle, scrub the preview, and say "Looks good, render it." The copilot confirms first, renders the MP4, and you approve it into a shareable post .

The Trackpad Editor chat dock with a hunt prompt typed: 'Hunt 9:16 clips about cold plunges from this podcast URL, ~5 clips around 3 seconds, no on-screen text', and a confirmation that a clip scan has started.
Step 2 — one chat message kicks off an AI clip hunt (/clips/scan) against a long-form source URL.
The cloud-status pop-panel showing a running clip-scan task with a progress indicator, alongside a grid of freshly hunted 9:16 raw clips in Library → Raws.
Steps 3–4 — the hunt runs async in the cloud-status panel; finished clips land in Library → Raws.
Note /clips/scan is for long-form sources (podcast, VOD, YouTube, TikTok, IG, X), and target_duration_sec is a soft range — asking for 3s yields roughly 2–4s clips. Because hunts are async, kick them off and keep working; the cloud-status panel tells you when they're done. Importing a file you already have should use import_only — the scan path is specifically for AI mining.
What it costs A hunt bills AWS compute (cents) and runs its AI tagging on your own keys; the raws become reusable footage — the "raw clips" paintbrush . Swapping them into scenes generates no AI video, so the whole remix stays in the cheap band; cloud renders start at $0.01.

Where to go next

You've mined long video into a fast remix. Go deeper on sourcing, or step up to replacing the on-screen actor with a generated one:

Hunt your first clips now

Fork a remix template, hand the AI a long podcast or YouTube URL, and let it mine the best short clips for you. Cloud renders start at $0.01.

Open the Discover feed