Critic

QA gate for video and image content. One endpoint — POST /v1/watch — with different prompts for different jobs.

Quality Check

Score content on a pass/fail basis. Catches bad hooks, visual artifacts, weak CTAs, and audio issues.

POST /v1/watch {
  "mediaUrl": "[MEDIA_URL]",
  "prompt": "Score quality 1-10. List issues with severity for: face consistency, artifacts, motion, audio sync."
}

Returns structured JSON with score and issues. If score is below threshold, regenerate.

Verification

Check that generated content matches the intended description. Compares the video against the original spec — did the hook land? Is the setting correct? Does the dialogue match?

POST /v1/watch {
  "mediaUrl": "[MEDIA_URL]",
  "prompt": "Does this video match this description: [ORIGINAL_DESCRIPTION]. Compare and list discrepancies."
}

Deep Analysis

Full content breakdown for strategy or debugging. Returns hook text, format classification, pacing notes, transition types, CTA analysis, and suggested tags.

POST /v1/watch {
  "mediaUrl": "[MEDIA_URL]",
  "prompt": "Analyze this video. Return: hook text, format, pacing, transitions, CTA, tags."
}

Key Behaviors

• Run a quality check on every video before posting. No exceptions.
• Verification is optional — use it when generation quality is inconsistent.
• Deep analysis is for strategy and debugging — don't run it on every video.
• If a video fails quality check twice, change the prompt or model before retrying.
• 10 credits per watch call.