Video Generation

AI Video Generation

Help users create and promote AI-generated video content.

Rules:

• Ask what they want: text-to-video, image-to-video, or video editing
• Check provider files for setup: runway.md, luma.md, kling.md, pika.md, stable-video-diffusion.md, replicate.md
• Check api-patterns.md for async polling, backoff, and caching
• Check promotion.md for distribution and marketing strategies
• Start with lowest resolution/duration to validate prompt before scaling

Platform Selection

Prompting for Video

• Describe motion explicitly — "waves crashing", "walking toward camera", not just scene descriptions
• Include camera direction — "camera slowly pans left", "dolly zoom", "tracking shot"
• Front-load subject — models weight early tokens heavily
• Under 200 tokens — longer prompts dilute motion guidance
• Avoid impossible physics — results look uncanny without proper setup

Image-to-Video

• Source image quality matters more than prompt — artifacts amplify
• Match aspect ratio BEFORE generation — post-crop destroys quality
• Use images with implied motion (mid-stride, flowing hair)
• Remove text/logos — they warp unpredictably

Aspect Ratios

• 16:9 — YouTube, standard horizontal
• 9:16 — TikTok, Reels, Shorts (vertical mandatory)
• 1:1 — Instagram feed, flexible
• Generate in target ratio — never crop after

Cost Control

• Start with shortest duration (2-4s) to validate
• Preview at lowest resolution (480p/720p)
• Extend good clips rather than regenerating
• Download immediately — signed URLs expire
• Free tiers: Pika (150/mo), Kling (66/day), Luma (30/mo)

Failure Modes

• NSFW false positive — rephrase or switch platform
• Face morphing — add "consistent face" to prompt
• Text always broken — add in post-production
• Hands/fingers wrong — frame to minimize visibility
• Quality degrades >8-10s — plan for cuts

Current Setup

Projects

Preferences

Empty sections = not configured yet. Check providers.md for setup.