Structure mode is a setting in ConrolNet that uses something called “canny mapping”, which pulls out a very detailed visual outline of every edge of a Reference Image it can detect
Structure mode maintains the most detail of the original reference image of any ConrolNet function.
What are good examples of reference images?
Structure mode is incredibly versatile, and any image can be considered a good image. What is more important than the image itself is how a designer intends to use the image.
Structure mode interprets its structure mode map to maintain as many details as possible, other than color and shading.
This means a couple things:
- Structure mode can be used with even an overfit or underfit model, as it will override the majority of the overfit failure users experience (repeated dataset images, warped lighting.)
- Structure mode will also ignore elements of the prompt that are in direct contradiction with the subject of the reference image. (see the jeep example below)
Feature highlights
Every mode in ControlNet has different features. We’ve shared the primary feature highlights below:
High coherence to detail
As mentioned earlier, structure mode pays the most attention to the details it has been provided in the reference image. We can see an example of this below with the Jeep.
If an image of a jeep style car is present in a reference image, the output will be very coherent and detailed, reimagined in the style of the model being used.
When presented with prompts that contradict that overall shape and design of a jeep, Structure mode instead uses that information to influence the color and general style of the output, which remains as a jeep structurally.
Useful for reskinning
Structure mode is possibly the most useful for reskinning an object, character or world in a different style. Designers need not worry about important structural aspects of their character being lost in interpretation, as structure mode will only change details such as color, style, shading, and depth of field.
Conclusion
Structure mode is a very powerful tool in ControlNet that adheres to all the original linework and edges in a reference image. As a reminder, a few important points to think about when you are choosing your reference image and prompt:
- Consider your use case - do you want your output to be incredibly similar in likeness to your input image?
- Compare your reference image to your prompt, and confirm they are not in conflict with one another.
Thanks for reading, and enjoy creating with structure mode!