What problem does it solve? AI video generations on Higgsfield often fail, drift, or ignore prompts, and without a systematic process users burn paid credits re-rolling blindly. This Skill provides structured diagnosis, verdict-based triage, and escalation rules so every failed take leads to a targeted fix instead of guesswork. ## Core Features & Use Cases - Symptom-to-Fix Tables: Map common problems (face inconsistency, dead camera moves, ignored prompts, static image-to-video, blocked dark content) to root causes and concrete prompt or input fixes. - Take Triage & Retry Ladder: Assign every delivered take one of five verdicts (keep, fix-in-post, edit, re-roll, rewrite), enforce one-variable-per-retake discipline, and escalate through a four-rung retry ladder that terminates after three paid attempts. - Outcome Logging & Vision Diagnosis: Log confirmed fixes to a persistent learning memory via scripts, and optionally classify rejected stills against a reject_reason enum with human confirmation. - Use Case: A Kling 3.0 Motion Control clip comes back with a drifting face. Instead of re-rolling, you consult the failure table, identify the character image framing as the cause, re-upload a full head-and-body reference, and log the confirmed fix for future sessions. ## Quick Start Ask the assistant to diagnose why your last Higgsfield generation failed or looked wrong, describing the symptom and the prompt you used.
