Creator Screening Skill

Automated creator/influencer screening powered by Memories.ai V2 Video Understanding API.

Parameters

Parameter	Default	Description
INLINECODE0	5	Number of top videos to analyze per creator
INLINECODE1

30 | First N seconds of each video to analyze | | platforms | instagram | Supported: instagram, tiktok, youtube | | analysis_mode | mai | mai (visual+audio AI analysis) or transcript (audio-only fallback) | | framework | default | Screening framework to apply. See references/frameworks/ | | output_format | discord | discord, pdf (Google Doc→PDF), or json | | batch_size | 10 | Max creators per batch run |

Quick Start

CODEBLOCK0

Workflow

Step 1: Parse Input

Accept creator URLs in any format:

• - Profile: INLINECODE14
• Individual reel: INLINECODE15
• YouTube: https://www.youtube.com/@channel or INLINECODE17

Step 2: Get Profile & Video Metadata

Memories.ai V2: INLINECODE18

CODEBLOCK1

Returns per video ($0.01/video):

• - Owner profile: username, fullname, followers, verified, profilepic
• Video stats: views, play_count, duration, caption, comments, dimensions, audio info

Step 3: Video Understanding (MAI)

Memories.ai V2 MAI: INLINECODE19

This is the core analysis step. MAI provides:

• - Visual scene descriptions: lighting quality, framing, environment, clothing, production value
• Audio transcription: speech-to-text with timestamps
• Content understanding: topic classification, delivery style, structure

CODEBLOCK2

Each video returns visual + audio AI analysis. Use this to evaluate:

• - Section 2.1: Look & Feel (lighting, environment, framing) — from visual scenes
• Section 2.2: Audio Quality (clarity, echo, consistency) — from audio analysis
• Section 3.x: Delivery & Content (structure, fluency, maturity) — from transcript text
• Section 4: Positioning (tone, energy, brand safety) — from combined analysis

Fallback: If MAI is unavailable, use --mode transcript for audio-only analysis:
CODEBLOCK3

Step 4: Apply Framework

Score against the selected screening framework:

CODEBLOCK4

Frameworks live in references/frameworks/:

• - cac-crusher.md — CAC Crusher Creator Screening Framework (Talking Head + Skit categories)
• INLINECODE23 — Generic quality screening (5 dimensions, weighted scoring)
• INLINECODE24 — Template for creating custom frameworks

Step 5: Generate Report

Output per-creator screening cards with:

• - Profile stats (followers, verified, engagement)
• Per-section scores (PASS/FAIL/FLAG)
• Transcript excerpts as evidence
• Visual quality notes from MAI
• Final verdict (APPROVED / REJECTED / CONDITIONAL)

Setup

Set the following environment variables before use:

CODEBLOCK5

Get your API key at https://api-tools.memories.ai

Memories.ai V2 API Reference

All endpoints use: Authorization: <API_KEY> header (no Bearer prefix).
Base URL: https://mavi-backend.memories.ai/serve/api/v2

Endpoint	Method	Use	Cost
INLINECODE27	POST	Profile + video stats	$0.01/video
INLINECODE28

POST | Visual + audio AI analysis | ~$0.11/video |
| /{platform}/video/transcript | POST | Audio transcription only | ~$0.01/video |

Platforms: instagram, tiktok, youtube, INLINECODE33

Request body: {"video_url": "...", "channel": "rapid"}
MAI response: {"data": {"task_id": "..."}} (async, results via webhook)

URL format: Instagram must use /reel/SHORTCODE/ (not /p/ or /reels/)

Error Handling

• - Add 0.5s delay between API calls
• Retry failed requests once, then skip
• If MAI webhook not received, fall back to transcript mode
• Always normalize Instagram URLs: /p/ → /reel/, /reels/ → INLINECODE42