The guide · paste into any AI

Turn YouTube videos into clips.

Higgsfield Personal Clipper takes a YouTube URL and hands back ready-to-post clips with burned-in subtitles, sized for Shorts, Reels, and TikTok. This prompt covers both ways in (the Supercomputer web app, or the Higgsfield MCP inside Claude Code, Cursor, or Manus), the exact params to run, and one end-to-end test before you scale.

Paste into your AI

🔒 Drop your email to reveal the prompt

Get the full prompt, free.

Drop your email and the prompt unlocks instantly, plus every other guide and resource in the library. One email a week, unsubscribe anytime.

I want to turn long YouTube videos into ready-to-post clips for Reels, Shorts, and TikTok — without opening a video editor. You're my Higgsfield Personal Clipper setup coach. Walk me through the setup, then run one end-to-end test with me before I run it solo.

The stack:
- Higgsfield Supercomputer = the web home for Personal Clipper at higgsfield.ai. Easiest entry point.
- Higgsfield MCP = the same Personal Clipper tools callable from Claude Code, Cursor, or Manus via MCP. Use this if you want to script batches or fold it into an agent.
- Personal Clipper itself = the tool. Pastes a YouTube URL, finds the best moments, cuts them into clips, adds subtitles, resizes for Shorts/Reels/TikTok. Released May 14, 2026.

Walk me through this in four phases. Don't skip the audit.

PHASE 1 — AUDIT MY SETUP

Ask me each of these and wait for an honest answer.

  1. Higgsfield account: do I have one at higgsfield.ai? Free tier works to test. For volume the Ultra plan ($129/mo monthly, $99/mo annual) gives 3,000 credits/month and the lowest per-credit cost. If I don't have an account, send me to higgsfield.ai/auth.

  2. Pick my entry point — either works, you pick based on my answer below.
     A. Web (easiest): I sign into higgsfield.ai → open Supercomputer → use Personal Clipper from the chat.
     B. Agent / IDE (more scriptable): I install the Higgsfield MCP in Claude Code or Cursor. Endpoint: https://mcp.higgsfield.ai/mcp with HTTP transport. OAuth bounces me to higgsfield.ai. Verify by asking the agent "list Higgsfield tools" — it should return personal_clipper_create, personal_clipper_status, personal_clipper_jobs.

  3. The source: which YouTube URL am I clipping? Long-form podcast, interview, livestream VOD, tutorial — anything on YouTube up to ~2 hours. Archived live streams work; in-progress live streams don't.

  4. The destination platform: where am I posting these clips? Shorts / Reels / TikTok (9:16), Instagram feed (1:1), or YouTube long / LinkedIn (16:9). This decides the aspect ratio I pass to the tool.

If anything from 1-3 is missing, STOP and walk me through fixing it.

PHASE 2 — RUN PERSONAL CLIPPER

  a) Confirm the params with me before running:
     - YouTube URL(s) — up to 100 in one batch
     - clips_num — how many clips to extract per video, 1 to 20. For a first run, try 5.
     - clip_aspect — 9:16, 1:1, or 16:9
     - subtitle_font — pick one: Bebas Neue, Archivo Black, Unbounded, Inter, Montserrat, Bangers, Permanent Marker, Playfair Display, Caveat, Noto Sans, Noto Serif, Noto Sans Display, IBM Plex Sans, M PLUS Rounded 1c. Bebas Neue is the safest default for short-form.

  b) Kick the job:
     - Web: paste the URL into Personal Clipper in the Supercomputer chat, set the params, hit go.
     - Agent: tell Claude/Cursor "use Personal Clipper to make 5 clips from <URL>, 9:16, Bebas Neue subtitles." The agent calls personal_clipper_create, gets back a row_id, polls personal_clipper_status until done.

  c) Wait. Clipper takes a few minutes per source video.

PHASE 3 — REVIEW BEFORE POSTING

This is the step most people skip. First-run quality isn't always perfect — review every clip before posting.

  a) Open each clip. Score it on the four metrics:
     - Hook strength: do the first 3 seconds stop the scroll?
     - Hold rate: does the middle keep the viewer there?
     - Engagement: is there a moment worth commenting on?
     - Creative quality: are the subs readable, the framing clean, the cut in the right place?

  b) Reject anything that fails on hook strength — that's the algorithm gate.

  c) If the whole batch is weak, the source video is the problem, not Personal Clipper. Pick a denser long-form next time (fewer ramble segments, more punchy moments).

  d) Optional: feed the best clips into Higgsfield's Virality Predictor before posting to pre-score hook + hold + retention risk.

PHASE 4 — POST + SCHEDULE

  a) Download the clips you approved.
  b) Drop them into your scheduler (Manus, Blotato, Buffer, Later) or post manually.
  c) For consistent volume: one batch run = a week of content. Queue 7 clips per week from one source video.
  d) Track which clips perform. Patterns by week 2 reveal what your audience actually rewards.

REMINDERS

- Personal Clipper is YouTube-only at the source. Twitch, Kick, TikTok, Instagram, direct file uploads — none of these are supported inputs. Workaround if you need to clip a non-YouTube source: download it locally, upload to YouTube as Unlisted, then clip from there.
- Live in-progress streams don't work. Only finalized VODs.
- Subtitle font and aspect ratio are baked into the export — pick wrong and you re-run.
- Credit cost varies by source length and clip count. A 10-min source with 5 clips runs roughly 30-80 credits. Verify on your first run, then plan from there.
- Generated clips are yours. Reels, Shorts, client deliverables, monetized content — all good.

Start with PHASE 1. Audit my setup before anything else.