Vary is a Scenario feature that lets you quickly generate two close variants of your images. It's useful for making subtle tweaks or testing creative alternatives while keeping the outputs aligned with your original intent before choosing the ones to further edit.
The Vary button is located to the top-right of the image thumbnail, in the generation screen. It will display the cost in credits. As you activate it, two new variants will be generated, depending on how the reference image was initially created:
Text-to-Image: Vary will trigger an Image-to-Image inference using your selected image as the reference, 25 influence, and applying the same prompt, and the same AI model
ControlNet: For images initially generated with ControlNet, Vary applies an Image-to-Image + ControlNet process, preserving the original ControlNet parameters unchanged.
Character/Style Reference: If your generation included a character or style reference, Vary will perform an Image-to-Image + Character/Style Reference inference.
Examine the generated variants and select the best fit, or keep iterating. You can also refine the approach and run more Image-to-Image inferences needed, adjusting the prompt or influence level to dial in more consistency or creativity.
An example of “Vary“: the original image below produced the two new variants above.
Generate an initial image using your preferred method (text-to-image, ControlNet, etc.). Apply “Vary“ by activating the “Vary“ button on the image thumbnail to generate two variations.
Review and select: examine the generated variants and select the one that best fits your vision. For further refinements, you may continue to iterate with the following approaches:
Apply “Vary“ again to your chosen variant
Set up specific Image-to-Image parameters and generate more variants with manually adjusted influence levels or prompts
Use other editing tools on your selected variant, like Retouch (Canvas), Enhance or Expand.
Vary is an efficient tool for fast iterations that helps you maintain consistency while exploring creative alternatives. Use it for small, incremental improvements rather than dramatic changes
When designing a character (ideally with a character model), Vary helps maintain consistent traits while exploring different poses or expressions variants.
Chain multiple “Vary“ operations to explore a range of options while staying on-style and on-model
Image-To-Image Generation: Scenario API Documentation - POST /generate/img2img
Image-to-Image + ControlNet: Scenario API Documentation - POST /generate/controlnet-img2img
Image-to-Image + Character or Style Reference: Scenario API Documentation - POST /generate/img2img-ip-adapter
Was this helpful?
Quentin