Summary

• Add per-input preprocessing rules through the existing C API, CLI, and server requests. Support resizing, cropping, padding, resampling, antialiasing, and Canny across all image targets, including indexed inputs.
• Decode images at their original resolution, merge presets with user overrides, and transform each input once before image or video generation. Keep init/mask geometry aligned and preserve downstream model processing and reference settings.

Related Issue / Discussion

N/A

Additional Information

N/A

Checklist

• I have read and confirmed this PR follows the contribution guidelines.