Learning how to achieve the results you want with Runway's generative tools is as easy as understanding what types of prompting the models understand best. You likely know how to best explain a situation, visual, or concept to someone you're close with; knowing how to explain a situation, visual, or concept to a Runway model is one of the most powerful (yet simple) ways to generate whatever you can imagine.

1. Runway prompts are not conversational

Other generative or AI-centric services (most commonly Large Language Models, or LLMs) you've encountered may encourage you to use a conversational manner of prompting in order to achieve what you want:

X "Can you make me a video of a dog jumping?"

X "Please show me a story about two friends."

With Runway prompts, these types of conversational additions will distract the model, and stop it from creating what you want to see. Instead, stick to descriptions of a visual. 

 

2. Runway prompts are not command-based

Similarly to Runway prompts being non-conversational, our generative models are not designed to follow commands. Some examples of what this may look like:

X "Add a dog to this clip of a field."

X "Make the candle go out."

While it might seem like you're providing directions, our models are looking for (and rendering from) descriptions of visuals (what is the camera looking at?), not commands.

 

3. Runway prompts apply to one visual item at a time

At present, inputting an entire screenplay or a series of shots will not output an entire multi-shot film or video. Rather, one prompt equals (creates) one shot (or one image, if you're using Text to Image). 

 

3. Runway prompts can be very technical

If you're a film buff, have prior production knowledge, have spent time on film sets, or just are generally able to do a quick internet search to see what different film terms mean (including but not limited to technical lighting terms, specific camera specifications, and lens types/effects), experimenting with technical film terms can provide you with fantastic results. For example, here's one prompt that regularly creates strong, cinematic photorealism by using technical film terms:

"[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"

 

4. Runway generative video prompts do not yet understand negative prompts

Negative prompts are prompts common to generative AI models in which a user supplies descriptive traits that they do not wish to have incorporated in the results. Here are some examples of negative prompts:

X "Bad composition."

X "Out of focus."

X "Poorly lit."

With the exception of Text to Image having a separate input box for negative prompts, our generative video models currently do not understand negative prompts, so please don't include them.

 

5. Runway prompts are best when describing what the "camera" is capturing

To generate anything you can imagine (and more), learning to describe a shot visually is the absolute best way to master Runway prompting.