Readme
Gen-4 Image with references
A reference-based image generation model that maintains character and location consistency across generations.
When attributing to Runway, please use “Powered by Runway” and link to runwayml.com from the user interface.
Overview
Gen-4 Image with references enables generation of consistent characters and locations using 1-3 reference images. The model preserves visual identity while allowing transformation of lighting, poses, settings, and styling through text prompts.
Core Features
- Character preservation: Maintain facial features and identity across different scenes
- Location consistency: Generate multiple angles and shots of the same environment
- Multi-reference support: Use up to 3 reference images per generation
- Conversational prompting: Natural language descriptions with iterative refinement
Optimal reference images
- Natural, even lighting (avoid harsh shadows)
- Neutral expressions for maximum flexibility
- Moderate to high quality
- Clear subject visibility
Usage Patterns
Single Reference
Use text prompts to describe transformations while preserving character identity.
@character_name wearing a leather jacket in a cyberpunk city
Multi-Reference
Combine character and scene references for precise control.
@character sitting in @location with dramatic lighting
Scene Generation
Create consistent environments from different angles.
show the castle courtyard from above
wide shot of the same location at sunset
Advanced Workflows
- Iterative refinement: Use generated outputs as new references to build complex scenes step-by-step.
- Separate pathways: Develop character and environment references independently, then combine.
- Element isolation: Generate specific clothing, poses, or lighting setups as references for final compositions.