FLUX Regional Prompting

[RyanJDick]

Summary

This PR adds support for regional prompting with FLUX.

Example 1

Global prompt: An architecture rendering of the reception area of a corporate office with modern decor.

QA Instructions

• Test that there is no slowdown in the base case with a single global prompt.
• Test image fully covered by regional masks.
• Test image covered by region masks with small gaps.
• Test region masks with large unmasked ‘background’ regions
• Test region masks with significant overlap
• Test multiple global prompts.
• Test no global prompt.
• Test regional negative prompts (It runs... but results are not great. Needs more tuning to be useful.)
• Test compatibility with:
  • ControlNet
  • LoRA
  • IP-Adapter

Remaining TODO

• Disable the following UI features for FLUX prompt regions: negative prompts, reference images, auto-negative.

Checklist

• The PR has a short but descriptive title, suitable for a changelog
• Tests added / updated (if applicable)
• Documentation added / updated (if applicable)
• Updated What's New copy (if doing a release after this PR)

[psychedelicious]

Disabling specific features based on the selected model was getting a bit tricky. For example, there are many paths to creating a regional guidance reference image (IP Adapter), so it's hard to "block" all of those when FLUX is the selected model.

I've made a few changes to ease the dev burden and improve the user experience.

When a layer is invalid, we might disable the Invoke button and or skip that layer during graph building. For example, a raster layer is invalid when it has no objects in its canvas state, and a control layer is invalid when its ControlNet model is for a different architecture than the main model.

I've consolidated the validation logic into helper functions that return a list of warnings. These warnings are displayed in the layer list items. For example, when the selected model is FLUX and a negative prompt is used for a regional guidance layer:

The same helper functions are used for the warning tooltips, Invoke button "readiness" checks, and graph builders.

When a layer is disabled, the icon is greyed out but still visible:

When there are no warnings, there is no icon.

The warnings should be the same the warnings in the Invoke button tooltip, when it is disabled - the validation logic was mostly copied from there.