AI Training offers a unique way to bring your own content to Runway's Text to Image tool. By inputting 15-25 images of your choice, you can develop a custom-trained model tailored to your specific needs.
Types of AI Training
Runway provides three types of AI Training to cater to all types of creative projects:
- Character Generations can be used for either humans or animals, and work best when supplied with portrait shots from the shoulders up.
- Style Generators are trained on the visual aesthetic of a set of images. They prioritize the "look" and feel of your creations, making them ideal for projects where a certain style is paramount.
- Object Generators are tailored for specific objects or things, perfect for product photography or showcasing non-living items in various styles and contexts.
Upgrading to a Standard, Pro, or Unlimited subscription tier grants you one free AI Training. This bonus is a one-time offer per initial subscription and does not renew with plan renewal or re-subscription. Additional trainings are available for purchase at 1000 credits each. It's important to note that deleting a training does not refund credits, so proceed with caution.
Best practices for training
To achieve the best results:
- Use at least 15 images; more images contribute to a better-trained model.
- Ensure photos are clear, focused, and feature the subject prominently.
- Avoid multi-figure shots for character models and strive for variety in context, lighting, and backgrounds.
- Choosing a unique model name like “custom[yourname]model” over simpler names helps avoid model confusion, ensuring a smoother and more personalized experience.
Model training typically takes between thirty minutes to an hour. This downtime is a perfect opportunity to take a break and plan out some of the prompts you might want to use with your model in Text to Image.
Using your trained model
To use your model in Text to Image:
- Go to the AI Training tab on the dashboard's lefthand menu.
- Click on your model’s thumbnail to open it in Text to Image.
- Adjust settings (ratio, resolution, outputs, style) and complete your prompt.
Each generation with your model costs 5 credits.
Switching between different trainings is seamless with the keyword switcher dropdown. You’ll need to keep the keyword in the prompt with the appropriate model selected for the training’s contents to be applied to your generation.
Currently, only one trained model can be used per Text to Image generation. This means you cannot currently train multiple models and incorporate them into the same final image using only Text to Image.