Lilly Tech Systems
Best Practices Advanced
Getting consistently great results from AI video generation requires a disciplined approach to prompting, credit management, quality assurance, and ethical considerations. This final lesson captures the key practices that separate experimental hobbyists from professional AI video creators.
Prompt Optimization
| Do | Avoid |
|---|---|
| Be specific about camera movement and subject action | Vague prompts like "make it look cool" |
| Include lighting and atmosphere descriptions | Relying on the model to guess mood |
| Reference specific visual styles (cinematic, documentary) | Contradictory style instructions |
| Keep prompts focused on one clear scene | Cramming multiple scenes into one prompt |
| Iterate: refine prompts based on outputs | Giving up after one generation |
Credit Management
- Start with shorter durations (5s) to test prompts before committing to 10s generations
- Use lower resolution for initial concept exploration, then upscale winners
- Track credit consumption per project to stay within budget
- Build a prompt library of proven prompts to reduce experimental waste
- Set team-level credit alerts to prevent overspending
Quality Control
QC Checklist: Before approving any AI-generated clip for production use, check for: temporal consistency (no flickering), anatomical accuracy (correct number of fingers, natural face), motion coherence (physics-defying movement), and artifact-free edges (no morphing or melting at frame boundaries).
Ethical AI Video Creation
- Never generate content that could deceive viewers into thinking it is real footage
- Disclose AI generation in video metadata and descriptions where appropriate
- Respect copyright: do not use copyrighted characters or identifiable real people without permission
- Follow Runway's content policy and community guidelines
- Consider the societal impact of your generated content
Staying Current
AI video generation is evolving rapidly. Runway ships major model updates multiple times per year. Follow Runway's blog and release notes, join the Discord community, and experiment with new features as they launch. Skills built on fundamentals (prompting, visual storytelling, pipeline design) transfer across model versions.