Video Prompt Generator

Problem addressed

When generating visual media, users often mix subject, style, composition, camera motion, lighting, and negative constraints into one natural-language sentence. The result varies across platforms. Video prompts are especially fragile because duration, camera motion, frame pacing, and physical movement need explicit treatment. This skill turns creative intent into structured, reusable prompts, reducing missing positive terms, weak negative prompts, platform syntax mismatches, and lost context during revision.

How it works

• Classify before decomposition: it detects image, video, tutorial, or reference-style requests, then loads the matching rules instead of applying image fields to video needs.
• Populate fields by priority: video focuses on V-SUBJ, V-CAM, V-ENV, and V-LIGHT; image focuses on subject, style, environment, and composition. P1/P2 fields activate only when the input is explicit enough, limiting keyword bloat.
• Build and validate prompts: it combines subject, spatial depth, motion, and lighting, checks contradictions such as “realistic but anime” or “daytime but fully dark”, then emits a positive prompt plus 10 to 25 negative terms.
• Adapt platform syntax: it adjusts weights, language, and formatting for MJ, SD, and Flux-style targets. Standard output stays concise; tutorial mode can explain the reasoning.

Boundaries

It is useful for prompt drafting and iterative refinement, not for model training, video rendering, or rights clearance. Ambiguous requests fall back to default modes and conservative defaults. Sensitive content, portrait rights, and high-risk copyright combinations are constrained or refused by safety rules.