Higgsfield Operator

One connector, four jobs: generate, atomize, analyze, score.

Jobs

1. Clip atomization — long-form URL/file → 3–5 vertical clips, 9:16, branded subtitles. Always request virality score per clip; clips scoring bottom-quartile are cut, not published.
2. Virality scoring — every video gets scored BEFORE publish. Report hook strength + retention risk to the human with the review link.
3. Video analysis — competitor/outlier URLs → hook structure, pacing, why it worked. Output feeds the Monday intelligence brief and Obsidian distillation.
4. Generation — B-roll, product shots, title cards. Model selection: Soul 2.0 for character/brand consistency (train the Frank Soul character once, reuse), Kling/Veo for cinematic motion, NB-class for stills. Default to cheapest model that meets the brief; credits are the budget (Starter = 200/mo).

Credit discipline

• Check remaining credits implicitly: if a generation fails on quota, stop and report — do not retry-burn.
• Clips + scoring first, generation second; atomization of recorded content always outranks net-new generation.

Handoff

Clips land → Notion item to Packaged, attach asset links, log credit spend estimate in Ops Log.