Runway’s model lineup changes quickly. The official getting-started guide currently recommends using the latest generation model shown in Tools and treating prompts differently for text-to-video versus image-to-video.
1. Open a generation session
In Runway, use Tools for direct generation control. Choose the current video model shown in the model selector.
2. Pick Text to Video or Image to Video
Text-to-video needs both visual and motion description. Image-to-video already gets appearance from the image, so the prompt should focus heavily on desired motion.
3. Describe one shot
Write subject motion, environmental motion and camera behavior. Keep the action physically understandable and avoid stacking several scene changes.
4. Generate a cheap/fast test when available
Use faster or lower-cost options for exploration when the interface offers them, then spend more on the direction that already works.
5. Iterate from the output
Use the finished generation as the next input/reference when appropriate, or change only the motion/camera instruction and compare.
One continuous shot. The subject [action]. The environment [secondary motion]. Camera [movement + framing]. Motion is natural and physically coherent. Keep subject identity and scene layout stable; no scene cuts.
Repeating the entire image description in image-to-video instead of describing motion; asking for multiple shots in one prompt; using negative language instead of stating desired motion; spending maximum credits before the concept is stable.
Finish check
A good first workflow produces one controllable shot and a prompt you can modify systematically, not a lucky one-off clip.