Imagine asking Firefly to generate a person. That person looks fine, but you'd really like it if they were looking in a certain direction, or at a particular object. By using this feature, the artist could mask a certain area (or the whole photo), input the rotational changes, and then preview generations.
This could be a very complex feature, but the basic idea is this:
- Load an image
- Mask the element you want re-oriented
- Interact with a context menu with XYZ rotational, focal length, and maybe QoL presets/swatches to select default parameters (ie. side, top, back views).
- Generate the output with multiple variations (as it currently does)
Bonus parameters to include
- Scale (ie. the difference between a pie on your plate and a pie the size of a city)
- Cohesion (ie. how much should the output look like the original reference)
- Directional Lighting
AdChoices