Generative models interpret your prompt and references rather than reproducing them exactly, so a result that drifts from what you pictured isn't an error. Getting closer usually takes a few rounds of adjusting what you give the model.

Knowing what each generation costs

Each generation uses credits, so a few habits keep the cost of iterating down:

• Check the model's rate — credit rates vary by model, and some use more credits per second of video than others; What are my credits worth? lists the rate for each model
• Have Agent ask first — in Agent settings (bottom right corner of the prompt panel), set the generation mode to Ask before generating media; Agent then shows its plan, model and estimated credit cost for each step and waits for your go-ahead before generating (see Creating with Runway Agent)
• Iterate at a lower resolution — generate at 720p while you refine, then upscale only the version you're keeping (see Upscaling videos and images)
• Optimize Agent for cost — under Optimize generations for…, choose Cost and Agent prioritizes lower credit usage when it picks a model
• Use Tool mode — select Tool in the left sidebar to choose the model, duration and settings yourself before each generation

Getting results closer to your idea

Most of what changes a result is the prompt, the references and the settings you choose. These guides cover how each one shapes the output:

• Introduction to prompting
• Image to video prompting guide
• Camera terms, prompts, & examples
• Prompt library

If you're working in Agent, you can also ask it to diagnose and fix a prompt that isn't working. For hands-on creative guidance, join our Discord community, where moderators and experienced Runway users can help with prompting techniques and workflow tips.

FAQ