Image Generation Step
Generate and edit images and variations
AirOps's Image Generation Step offers three powerful functions:
Generating Images
Editing Images
Generating Variations
We'll dive into each, and the tools that power them.
Generate Image
When generating images, a Workflow Input isn't necessarily required. Instead, you can utilize the "Generate Image" action shown below:
DALL·E 2 allows you to choose how many images to generate.
Stable Diffusion allows you to include a negative prompt (i.e. write in what you would like the image to avoid).
Supported AI Models for Image Generation
AirOps offers 4 AI models at this time:
DALL·E 3 -- introduces novel techniques enhancing image synthesis, allowing users to push the boundaries of creativity
DALL·E 2 -- strikes a balance between model complexity and performance, making it versatile for various workflows
Stable Diffusion XL v1.0 -- brings stability to the forefront of image generation, ensuring reliable and consistent results
Stable Diffusion XL v0.9 -- an earlier version with advancements in stability and performance, potentially a better option for more performance-constrained environments
Supported Image Sizes for Image Generation
The supported image sizes will vary depending upon which model you choose.
DALL·E 3
1024x1024
1024x1792
1792x1024
DALL·E 2
1024x1024
512x512
256x256
Stable Diffusion XL v1.0/v0.9
21:9 (1536x640)
16:9 (1344x768)
3:2 (1216x832)
5:4 (1152x896)
1:1 (1024x1024)
4:5 (896x1152)
2:3 (832x1216)
Supported Image Format for Image Generation
DALL·E 3, DALL·E 2
PNG
JPEG
WEBP
Stable Diffusion XL v1.0/v0.9
PNG
Edit Images
The Image Generation Step also allows you to edit images. To do this, you must have an input in the form of either:
An advanced "File Media" Workflow Input variable
A publicly accessible URL with a .png endpoint
Additionally, you must provide a similar input for an Image Mask URL (an image of the same size, with fully transparent areas to mark where the image should be edited).
Editing images is currently only supported with DALL·E 2.
In the example below, we will use the same public URL (with a transparent background) for our Image URL and our Image Mask URL.
Generate Variations
As a third option, the Image Generation Step allows you to generate variations of an existing image. Similar to editing an image, you are required to have an input in the form of either:
An advanced "File Media" Workflow Input variable
A publicly accessible URL with a .png endpoint
Generating variations is currently only supported with DALL·E 2.
If you choose to generate multiple images, the output will be surfaced in a JSON format with the accessible URL's for each distinctly generated image.
Last updated