Image Upscale

Problem

Enlarging low-resolution bitmaps often exposes missing detail: feathers, textures, and edges become soft, while plain interpolation can introduce jaggies. Design assets destined for print also often lack consistent DPI metadata.

How It Works

The skill combines AI upscaling with post-processing:
- Parameter check: confirm the target scale or pixel size and the required DPI.
- Input handling: supports PSD, PNG, JPG, and JPEG; PSD files are flattened and converted to RGB.
- Real-ESRGAN: uses RealESRGAN_x4plus.pth for 4x upscaling, rebuilding texture and edge detail; large images are processed in tiles to reduce memory pressure.
- LANCZOS: if the target exceeds 4x, interpolation reaches the final size while keeping the output consistent.
- Output control: writes the requested DPI, produces a consistently named PNG, and supports single-file or batch directory processing with existing files skipped.

Scope

It is best suited for design drafts, illustrations, and asset upscaling for print preparation, not for inventing missing original detail. Results beyond 4x rely on interpolation and depend on source quality. Local dependencies include PyTorch, Pillow, NumPy, and requests, with GPU acceleration available for inference.