Firefly Image 3 Model
The Firefly Image 3 Model offers an improved generative Al image model for faster ideation and content creation.
Overview
This document provides an overview of the changes and new APIs introduced for the Firefly Image 3 Model.
Specific benefits include:
- Improved generation quality with better depiction of details and more photorealistic results.
- Greater variety in generated outputs from a single prompt for photos, art, or flat illustrations.
- Improved ability to take reference images into account, both for stylization and for structure matching.
- Better understanding of prompts and compositions of complex scenes and text rendering.
- Improved image variations with varied options for composition, mood, and lighting.
See this link for more details on what's new.
New APIs
The following new APIs were introduced:
- Generate Similar Images: Pass in a source image to use as a reference for generating similar image results.
- Generate Object Composite: Upload an image (with or without mask), such as a product photo, and utilize a text prompt to generate a seamlessly composited scene featuring the product. Additionally, provide optional parameters to influence the style and content of the background. See this link for more specific details.
Updated APIs
The following existing APIs were updated to use the new Image Model 3, and have new and/or renamed parameters:
The
n
parameter has been renamed tonumVariations
.The
styles
object parameter was renamed tostyle
and contains renamed child parameters as well. You can still supply a reference image to be used, but it should now be passed asurl
(for a presigned image URL), oruploadId
for an uploaded image within theimageReference.source
field:Copied to your clipboardstyle": {"presets": ["bw"],"imageReference": {"source": {"url": "https://example.com/","uploadId": "string"}}"strength": 50,}The
photoSettings
parameter has been removed.The
locale
parameter has been renamed topromptBiasingLocaleCode
.
The
n
parameter has been renamed tonumVariations
.locale
has been renamed topromptBiasingLocaleCode
.The
image
parameter now has asource
and amask
that can be provided as sub-parameters in the form of either anuploadId
of an uploaded image (ie:image.source.uploadId
orimage.mask.uploadId
, or a url for a presigned URL (ie:image.source.url
orimage.mask.url
):Copied to your clipboard"image": {"source": {"url": "https://example.com/","uploadId": "string"},"mask": {"url": "https://example.com/","uploadId": "string"}}A new
placement
object parameter can be specified, withinsets
andalignment
child parameter values that can be provided. See the placement section below for more details.
The
n
parameter has been renamed tonumVariations
.locale
has been renamed topromptBiasingLocaleCode
. - Support for anegativePrompt
parameter has been added.The
image
parameter now contains both thesource
parameter for theimage
input, and themask
parameter for the mask to use. (The previous version of the API hadmask
at the same level as theimage
parameter). Please note, for both, you can supply either aurl
(for a presigned URL – ie:image.source.url
orimage.mask.url
) or anuploadId
(for an uploaded image – ie:image.source.uploadId
orimage.mask.uploadId
):Copied to your clipboard"image": {"source": {"url": "https://example.com/","uploadId": "string"},"mask": {"url": "https://example.com/","uploadId": "string"}}
See this link for more details about the capabilities available for expanding and filling images with Firefly Image Model 3.