Coconote
AI notes
AI voice & video notes
Export note
Try for free
Runway's Gen 3 AI Model for Text to Video
Jul 4, 2024
Runway's Gen 3 AI Model for Text to Video
Overview
Company
: Runway
Model
: Gen 3
Previous Models
: Gen 1 and Gen 2
Key Feature
: Converts text prompts into video clips
New Feature
: Lip sync - characters can read scripted text using AI voices
Access and Pricing
Website
: runwayml.com
Availability
: Open to everyone
Subscription
: Starting at $15/month
Credit System
:
10 credits per second of video generation
5-second video: 50 credits
10-second video: 100 credits
Using Gen 3
Website Navigation
Visit runwayml.com and create an account
Click 'Try Gen 3' (currently in Alpha)
Log in with your account
Navigate to the Gen 3 page
Prompting Guide
Prompts often consist of three parts:
Camera movement
Scene description
Additional details (e.g., lighting)
Example: “Low angle static shot of a woman wearing orange”
Manual descriptions required for camera settings, lighting, etc.
Use AI tools like ChatGPT or Claude to help generate prompts
Creating Videos
Enter the prompt into the specified box
Choose video length (5 or 10 seconds)
Generate video (time varies between 60-120 seconds)
Download or save the video once generated
Settings and Options
Watermark
: Option to remove Runway watermark
Seed Setting
: Maintain consistency in generated frames if enabled
Custom Presets
: Save frequently used prompts for easy access
Results and Limitations
Consistency
: Mixed results - one out of three might be satisfactory
Experimentation
: Necessary due to variability in output
Lip Sync Quality
: Basic, better alternatives available (like HeyGen)
Example Generations
Submarine in Deep Ocean
: Generated clip looks realistic
Detailed Prompt
: May result in awkward motion or unrealistic features
Additional Resources
Course
: “Making Movies with AI” on Skill Leap
Combines ChatGPT, MidJourney, Gen 2, and Gen 3 to create films
Includes a free trial with 22 video tutorials
Conclusion
Gen 3 is a promising yet developing tool in text to video conversion with significant potential improvements ahead.
📄
Full transcript