The image node allows you to generate images from prompts and other images using models provided by Fal.ai.
The image node is particularly useful for quickly illustrating concepts and ideas. Runchat provides access to the stable-diffusion-turbo model for free for the time being, but if you connect a Fal API key then the Image node will automatically add several other much higher quality models.
<aside> 💡
Check out the Image Models Guide for a comparison of the different default image generation models.
</aside>
The image node will generate one image for each input, just like every other node in Runchat. This means if you provide a list of prompts, the node will generate a list of output images in parallel. The generated images are displayed in a grid in both Editor view and App view. The image node has two parameters, Prompt and Image. The image parameter is only displayed if you use an image-to-image model and set an Image Strength of over zero.
Use the prompt input to describe what you want your image to look like. Newer, larger models are better at following prompts than older, smaller and cheaper models. When using image-to-image models your prompt is still used to describe the qualities of the overall image and can not be used as “direction” for how to change the image (”change the shoe to red” etc). For these use cases, consider using Fal models that support this kind of control using the API Node.
Loosely keeping the ‘form’ of an image while changing the material using image to image
To enable the Image input, select an image-to-image model from the model settings. Then drag the Image Strength slider to a number larger than 0. You can connect any image to the Image input and it will be used as the starting image for the diffusion process. This means that the image model will gradually change this image to more closely match your prompt. When image strength is 0, the diffusion process starts from random noise. When image strength is 1, the diffusion process starts from the reference image. A value of 0.5 is 50/50 random noise and your reference image.
<aside> 💡
The Image node tries to generate images as quickly as possible and does not run many iterations of diffusion. Setting high values for image weights therefore tends to result in only very small changes to the images. If you need more control over image generation, consider calling the same Fal models using the API Node.
</aside>