Product Updates
Introducing Control Layers, a powerful UI for applying regional prompts
May 9, 2024
Today, we’re excited to announce the launch of Control Layers, a powerful, new tool for applying different types of guidance – like compositional structure, textual prompts, or visual references – specifically where you need them in your images. 

Start with a blank canvas or basic sketch that provides some structural guidance (using ControlNets). Then, add layers to your canvas with text-based or image-based guidance (using IP Adapter) to infuse details into distinct, masked areas of your image with pinpoint accuracy. 

You can watch the full release video here.

With Control Layers, users can choose to incorporate a variety of different layer types to their image:

  • Regional Guidance
    • Regional Guidance (Text): With text-based regional guidance, users paint a masked area on their image or canvas and provide a text prompt. The AI model incorporates that prompt only in the area of the image defined by the user.
    • Regional Guidance (Image): With image-based regional guidance, users paint a masked area on their image or canvas and upload an image. They can then select how they want their AI model to understand that image (eg. incorporate just the style, just the compositional components, or all of the above), and how strong they want the image to influence the area they have defined.
  • Global Guidance
    • ControlNet (Structural Reference): With Global ControlNet, users can upload an image and the AI model will identify the structural elements of the image and create a global layer that follows the structural components of your image. This layer can be used as a base layer, with Regional Guidance added to areas of the image.
    • IP Adapter (Style/Compositional Reference): With Global IP Adapters, users can upload a reference image and the AI model will identify the style and composition of the image. The user can then choose how they want the model to incorporate those elements into the generation process (eg. style only, composition only, or both). 

Control Layers builds on Invoke’s focus on building creation tools that provide artists and creators with the highest level of control in their creation process. By combining different Control Layers, users will have a new level of creative control that hasn’t been possible in any other generative AI platform to date. 

We are releasing Control Layers for both our freely-available Community Edition and Professional Edition.

We’ve been testing this product for the past month and have come up with some learnings and best practices for effectively combining and using Control Layers here.

As you explore, we’d love to hear what you are learning. Please join our Discord and share your findings.

Happy Invoking!