Critic

Use Vidjutsu Watch for quality analysis. The production API base is https://api.vidjutsu.ai; raw calls require bearer auth. Watch is synchronous, limited to 50 requests/day, and included in the subscription.

Quality check

vidjutsu watch --mediaUrl "$MEDIA_URL" --prompt 'Return only JSON: {"score": number, "issues": [{"severity": "minor"|"major"|"critical", "description": string}]}. Score face consistency, artifacts, motion, and audio sync.'

The API response envelope is:

{"response":{"score":8,"issues":[]}}

Read model output from response; do not expect score at the top level.

Verification

vidjutsu watch --mediaUrl "$MEDIA_URL" --prompt 'Return only JSON describing whether this video matches: <description>. Include matches:boolean and discrepancies:string[].'

Deep analysis

vidjutsu watch --mediaUrl "$MEDIA_URL" --prompt 'Return only JSON with hook, format, pacing, transitions, CTA, and tags.'

Rules

• Supply a fetchable media URL, not an Instagram or TikTok page URL. Stage

social sources first with the Vidjutsu download endpoint.

• Parse and validate response against the schema requested in the prompt.
• If a video fails twice, change the generation prompt before retrying.
• Run quality QA on every final video before publishing.