Riverflow Reference-Based Super-Resolution

Riverflow Reference-Based Super-Resolution is an agentic detail repair capability designed to fix blurry, incorrect, or low-resolution details in an image by using reference images as ground truth.

It’s ideal for restoring things like: - product labels and packaging text - logos and brand marks - printed designs on objects (bottles, cans, boxes, apparel) - fine text and small UI-like elements inside a scene

What it does

Given: 1) an original image (the image that contains incorrect/blurred details), and
2) up to 4 reference images (the correct details),

the model will: - locate the best matching regions in the original image - super-resolve and correct those regions in-place - preserve the surrounding scene while improving detail fidelity

The model can fix up to 4 matched instances in a single run.

Inputs / Parameters

imageUrl (string, required)

The original image to repair.

This image can be large and does not need to match a standard aspect ratio.

superResolutionReferences (array, required)

A list of up to 4 reference images that represent the correct details you want restored.

Use references that clearly show: - the correct text - the correct artwork - the correct logo/label design

idempotencyKey (string | null, optional)

An optional key you can provide to safely retry the same request without duplicating work.

How matching works

The model attempts to find up to 4 best matches between the reference images and regions in the original image.

Examples: - 4 bottles in a scene that match 1 reference label → up to 4 fixes - 1 bottle each matching 4 different reference artworks → up to 4 fixes - No matching regions found → no changes applied

Pricing behavior (important)

Reference-Based Super-Resolution is priced per matched instance found at runtime.

That means: - the final cost is only known after submission - if 0 matches are found, the additional cost is $0 - up to 4 matches may be applied in one request

Best practices

To get the best results: - Use clean, high-resolution reference images - Make sure references are cropped tightly around the detail you want restored - Avoid references with glare, motion blur, or heavy perspective distortion - Provide multiple references only when they represent different target details

Common use cases

• Fix unreadable label text on bottles/cans
• Restore correct branding on packaging
• Correct small printed product details in lifestyle scenes
• Improve legibility of fine typography embedded in an image