Mastering Runway’s generative tools is all about understanding the type of prompts that the models respond to best. Just like you know how to explain a situation, image, or concept to a friend, learning how to convey these to a Runway model is a key (yet straightforward) way to bring your creative ideas to life.
1. Runway prompts are descriptive, not conversational
While many AI services, particularly Large Language Models (LLMs), may encourage a conversational style of prompting, this isn’t the case with Runway:
- 🚫 “Can you make me a video of a dog jumping?”
- 🚫 “Please show me a story about two friends.”
In Runway, conversational language can confuse the model, preventing it from producing the desired result. Instead, focus on describing the visual you want to create.
2. Runway prompts are not command-based
Just as they aren’t conversational, Runway’s models don’t follow direct commands. Here’s what that might look like:
- 🚫 “Add a dog to this clip of a field.”
- 🚫 “Make the candle go out.”
Although these sound like clear instructions, the models work best when provided with visual descriptions—what the camera should be looking at—rather than commands.
3. Runway prompts focus on one visual element at a time
Currently, inputting a full screenplay or a sequence of shots won’t result in a multi-shot video. Instead, each prompt generates a single shot (or a single image when using Text to Image).
4. Runway prompts can include technical details
If you have a background in filmmaking or production, or can look up film terminology (such as lighting techniques, camera specs, or lens types/effects), using technical film terms in your prompts can yield impressive results. Here’s an example of a prompt that often leads to cinematic, photorealistic outcomes:
“[SUBJECT] sharp focus, extremely detailed, photorealistic, RAW footage, 8k high resolution, RAW candid cinema, 16mm, color graded Portra 400 film, ultra-realistic, cinematic film, subsurface scattering, ray tracing, volumetric lighting.”
5. Runway generative video prompts don’t support negative prompts yet
Negative prompts, where users specify traits they don’t want to appear in the results, are not yet supported in Runway’s generative video models (though Text to Image does offer a separate input for them). Examples of negative prompts include:
- 🚫 “Bad composition.”
- 🚫 “Out of focus.”
- 🚫 “Poorly lit.”
For now, avoid using negative prompts with generative video models.
6. Runway prompts are most effective when describing what the “camera” captures
To unlock the full potential of Runway, focus on detailing what a camera would see in your desired shot. This approach is the most effective way to turn your creative vision into reality.
Explore other tips and tricks in our Blog.