Dream Machine: AI Video Generation
From the course Luma AI: 3D & Video from Photos
Dream Machine is Luma's AI video generation model, creating high-quality videos from text and image inputs.
Text-to-Video
Generate videos from text descriptions:
Prompt Structure
[Camera movement] [Subject] [Action] [Setting] [Lighting/Mood]
Examples:
- "Slow dolly forward through a misty enchanted forest at dawn, volumetric light rays filtering through ancient trees"
- "Aerial drone shot of a bustling Tokyo intersection at night, neon lights reflecting on wet streets"
- "Close-up macro shot of a butterfly emerging from its chrysalis, soft natural lighting"
Camera Controls
- Dolly: Forward/backward movement
- Pan: Left/right rotation
- Tilt: Up/down rotation
- Orbit: Circular movement around subject
- Zoom: In/out focal length change
- Crane: Vertical movement
Image-to-Video
Animate static images with AI:
- Upload a reference image
- Describe the desired motion and camera movement
- Dream Machine animates the image naturally
- Maintains the visual style and content of the source image
Use Cases
- Animate product photos for e-commerce
- Bring illustrations and artwork to life
- Create motion from AI-generated images
- Produce b-roll from stock photography
Generation Settings
- Duration: Typically 4-5 seconds per generation
- Resolution: Up to 1080p
- Aspect ratios: 16:9, 9:16, 1:1
- Keyframes: Set start and end frames for controlled motion
- Extend: Generate additional seconds to continue a video
Achieving Cinematic Quality
Tips for professional-looking results:
- Include camera movement descriptions for dynamic footage
- Specify lighting conditions for mood and atmosphere
- Use film terminology: "depth of field", "rack focus", "slow motion"
- Reference real-world camera types: "shot on ARRI Alexa", "35mm film look"
Named Example: Runway & Pika vs Dream Machine
When Luma AI launched Dream Machine in mid-2024, it immediately benchmarked favorably against Runway Gen-2 and Pika Labs in cinematic realism and camera motion control. Creative studios like A24 and MPC began exploring Dream Machine for pre-visualization work, using prompts like "slow push through an abandoned brutalist library, late afternoon golden light, 35mm anamorphic" to generate mood-reference clips for director pitches in under 5 minutes.
Cinematic prompt template:
"[Shot type + camera movement] of [subject], [action/state],
[setting], [time of day + lighting quality], [film aesthetic],
[lens type and characteristics]"
Example:
"Slow tracking shot following a lone astronaut walking across
a red Martian landscape, dust swirling around boots,
harsh midday light with long shadows, IMAX 15/70 film grain,
anamorphic lens flares"
Common Pitfall: Dream Machine generations with very fast camera movements (quick pans, handheld shakes) often produce motion blur artifacts and spatial inconsistencies. Start with slow, deliberate movements (dolly, slow orbit) and only introduce faster movement once you've confirmed your subject renders cleanly.
Key Takeaways
- •Dream Machine generates high-quality videos from text prompts with camera movement control
- •Image-to-video animation maintains source visual style while adding natural motion
- •Structured prompts with camera, subject, action, setting, and mood produce best results
- •Film terminology and camera references in prompts achieve more cinematic output quality
Progress is kept in this browser only.