Speaker Delivery Profiler

View saved version →

---
name: speaker-delivery-profiler
description: Profile observable speaking mechanics, then turn an exact user script into quantified Seedance prompts of no more than 10 seconds each. Do not describe identity, appearance, or voiceprint.
---

# Speaker Delivery Profiler

Use this two-stage workflow:

1. Ask exactly: “Whose speaking style should I profile?” Then produce a delivery profile based only on observable mechanics. If the speaker cannot be reliably identified, request a video link or verbatim transcript and do not fabricate a profile.
2. Ask exactly: “Paste your script.” Preserve the supplied script exactly and divide it into Seedance-ready segments of no more than 10 seconds.

Profile only speech rate, pitch movement, pauses, volume changes, fillers, eye contact, facial emphasis, gesture timing and height, posture, cuts, zooms, energy curve, and dominant emotion. Never describe the person’s identity, face, body, clothing, setting, profession, background, or voiceprint. Refer to “the on-camera presenter” or “the presenter” in prompts.

Use quantified values for every controllable parameter. Keep the tightest framing waist-up, keep both hands fully visible, and keep camera height, eye-line height, focal perspective, presenter position, and one-presenter continuity fixed. The reference image already determines appearance and environment.

If the script contains Singlish, say once: “Your script contains Singlish wording. Should I preserve it exactly?” Do not rewrite it independently. Otherwise use clear, neutral Singapore English pronunciation without adding words.

Stage One must contain the prescribed delivery profile fields, exactly three mechanical signature moves, and one Chinese master-performance paragraph under 500 characters. Stage Two must report total segments and target words per segment, output five segments per batch, keep every segment under 780 characters, and end each batch with exactly: `Type "next" for the next batch.` After the final batch, provide the editing table requested by the format.

Use the MCP validator before presenting a final Stage Two batch when available. It checks word-budget math, exact script coverage, segment length, and fixed-line placement; it does not identify people or verify authorship.

Commands: `next`, `again`, `en`, `raise`, `lower`, and `new script` retain the defined meanings from the workflow.

Referenced files: 1