Gen-3 Alpha vs Gen-2: What Actually Changed
The quality jump from Runway Gen-2 to Gen-3 Alpha is one of the largest generational improvements in AI video to date. The three most noticeable improvements:
- Temporal consistency — objects no longer morph between frames. A specific jacket stays the same jacket throughout a clip.
- Lighting and shadows — Gen-3 generates directionally coherent shadows and reflections. Gen-2 lighting was often flat or inconsistent.
- Camera control — prompting for specific camera movements (dolly in, arc shot, handheld) now works reliably. Gen-2 often ignored camera direction prompts.
Motion Brush
Motion Brush is Runway's interface for selective animation. You paint a region of a still image and specify the direction of movement for that region. The rest of the image stays static. This enables effects like:
- Animating only a waterfall in a landscape photo while keeping everything else still
- Adding hair movement to a portrait without distorting the face
- Making a flag wave without moving the building it's attached to
It is more predictable than full image-to-video generation because you are constraining the degrees of freedom the model works with.
Image-to-Video
Gen-3 Alpha accepts a still image plus a text prompt describing the motion, then generates video from it. The image anchors the starting frame, giving you compositional control that pure text-to-video cannot match.
# Via Runway API
curl -X POST https://api.dev.runwayml.com/v1/image_to_video \
-H "Authorization: Bearer YOUR_RUNWAY_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gen3a_turbo",
"promptImage": "https://your-image-url.com/scene.jpg",
"promptText": "Camera slowly pushes forward into the foggy forest",
"duration": 5,
"ratio": "1280:768"
}'
Act One: Facial Expression Transfer
Act One lets you record your own facial performance on a webcam and apply it to a generated character or existing footage. The model tracks landmarks from your face and retargets the expressions to the target character while preserving the character's appearance.
This is the pipeline that smaller studios use for animated character dialogue without motion capture hardware. Quality is strongest for frontal faces; profile views still degrade the tracking.
Pricing
Runway uses a credit system:
- Standard plan: 625 credits/month included
- Gen-3 Alpha: 50 credits per second of video generated
- At $15/month for Standard: 12.5 seconds of video before paying overage
- Additional credits: $0.05/credit ($2.50 per second of Gen-3 video)
For comparison, Sora on ChatGPT Pro ($200/month) includes 500 priority generations with no per-second pricing — better value for high-volume generation.
Comparison: Runway API vs Stable Video Diffusion Locally
If you need on-premise generation (for client data privacy or cost at scale), Stable Video Diffusion (SVD-XT 1.1) can be run locally on an RTX 4090. Quality is below Gen-3 Alpha but above Gen-2. At high volume (thousands of clips), local SVD becomes cheaper than Runway API credits.
For studios generating under 100 clips per month, Runway's managed API is faster to integrate and easier to maintain.
