ImgGen AI Image Upscaler vs WaveSpeed API

ImgGen's documented `factor` is a magnification instruction. WaveSpeed's `target_resolution` offers published 2K, 4K, and 8K choices. These are not interchangeable units. For each source size, calculate the intended output and choose the nearest appropriate WaveSpeed target. Do not copy `factor=4` into a field that expects `4k`. ImgGen also documents `enhanceFace` and `sharpen` switches. The standard WaveSpeed upscaler does not list those controls. If the old workflow depends on either, test the resulting pixels or retain a separate editing stage. No migration plan should imply hidden one-to-one controls.

The ImgGen example returns encoded image content in its result. WaveSpeed uses a prediction submission and result retrieval path. Adjust client code for asynchronous state, returned output location, error handling, and storage before changing production traffic. Keep a correlation ID that ties the source asset to the output and its review decision. Test with one product package bearing small typography and a reflective edge. Save both outputs from the same original; comparing files processed from different source sizes can create a false performance story.

For a migration pilot, record output dimensions, file format, edge changes, labeling accuracy, and reviewer time. A technically valid response can still be unsuitable for regulated copy or catalog markings. Have a fallback route when an image fails approval. Do not publish invented benchmark numbers from an unrun comparison. The image API collection is useful for exploring adjacent options, but each model has its own schema. Map only fields documented for the standard endpoint in the current integration.

If the pilot passes, migrate one class of assets at a time and keep the old output path until reviewers accept the new one. A queue can track individual predictions, as described in the bulk workflow guide. That queue belongs to your application. Log unexpected output sizes and inspect them before increasing traffic.