Real API outputs

AI-powered image processing behind one Rails 8 API.

Remove backgrounds, select people or objects, and transform images through simple API requests. Every example below uses a real API response so you can see what each capability actually does.

See what the API can doApplication source
BiRefNetKeep the subject, remove the background
MobileSAMSelect a person or object from a click or box
libvipsResize, crop and enhance predictably
Rails 8One API surface for the workflow
Original outdoor portrait Portrait after BiRefNet background replacement
↔
OriginalBiRefNet
Background removed and refined for a studio-ready portraitDrag to compare
Background removal

Keep the product. Remove the scene around it.

BiRefNet separates the main subject from its surroundings. The API can return a transparent PNG or place the isolated subject onto a new background.

Original chair in a room
Input

Original product photo

The chair is still surrounded by its original wall and floor.

Transparent chair cutout
BiRefNet

Transparent cutout

The wall and floor disappear while the chair silhouette remains.

Chair on a soft neutral background
Composite

Catalog-ready background

The isolated chair is recomposited onto a clean neutral backdrop.

Another object shape

The same background-removal API on a different object.

The hat has a wider curved outline, showing how the same endpoint handles a different shape and material.

Original hat photo
Input

Original hat photo

Large curved silhouette on a textured wooden background.

Hat transparent cutout
BiRefNet

Transparent hat cutout

A second object-removal case with a substantially different outline.

Hat on neutral background
Composite

Neutral editorial backdrop

Feathered output recomposited onto a clean neutral background.

Object selection with MobileSAM

Three animals. Three clear selection examples.

These examples separate the interaction from the result: the first image shows the prompt you give MobileSAM, the second highlights what the model selected, and the third is the raw black-and-white mask returned by the API.

Green point prompt placed on the puppy1 · Point prompt Puppy selected while the background is dimmed2 · Selection Raw MobileSAM point mask for puppy3 · API mask
Dog · Point

Choose a subject with one click

A single positive point tells MobileSAM which nearby object the user means. This is the simplest interaction for a click-to-select interface.

Blue bounding box around the cat1 · Box prompt Cat selected while the surrounding area is dimmed2 · Selection Raw MobileSAM box mask for cat3 · API mask
Cat · Box

Choose a subject with a bounding box

The box narrows the search region around the cat. This interaction fits crop tools, annotation UIs, and workflows where the user can drag a rectangle.

Green point prompt on the foreground white horse1 · Point prompt Foreground white horse selected while other horses are dimmed2 · Selection Raw MobileSAM point mask for foreground horse3 · API mask
Horse · Point

Select one animal from a busier scene

Several horses are visible, but the positive point targets the white horse in front. This makes the purpose of a prompt much clearer than an image containing only one obvious subject.

How to read these examples. The prompt image shows what the user provides. The selection preview is only a visual aid for this showcase. The black-and-white mask is the actual pixel-level MobileSAM output used by downstream code.
A human subject

Background removal and segmentation solve different problems.

The same portrait makes the distinction easy to see: BiRefNet changes the scene around the person, while MobileSAM identifies the pixels that belong to the person.

Original person portraitOriginal Person with original background replacedBiRefNet MobileSAM person maskMobileSAM mask
Person

One photo, two different AI tasks

Use background removal when you want a cutout or a new scene. Use segmentation when you need a mask that tells software exactly which pixels belong to the selected subject.

Predictable image transformations

Resize, crop, enhance, and export predictably.

These operations do not use an AI model. libvips performs repeatable resize, crop, color, sharpening, tint, and WebP transformations.

Pho originalOriginalPho enhanced720×720
Food

Social-ready square crop

Resize-to-fill plus restrained contrast, saturation, sharpening, and WebP output.

Hạ Long originalOriginalHạ Long cinematic960×540
Travel

Cinematic hero image

A 16:9 crop with subtle tonal enhancement for web hero placement.

Tower originalOriginalTower warmWarm
Architecture

Warm editorial treatment

Resize, contrast, restrained tint, sharpening, and WebP export in one request.

Explore the outputs

Open any result at full resolution.

Each item opens in a lightbox with zoom, navigation, and its original pixel dimensions.

Transparent cutoutBiRefNet · PNG Soft product compositeBiRefNet · feathered Hat cutoutBiRefNet · object Hat compositeBiRefNet · feathered Food enhancement720×720 · WebP Travel hero960×540 · WebP Architecture treatment960×540 · WebP Person background replacementBiRefNet · portrait Person segmentation maskMobileSAM · box Dog point selectionMobileSAM · point Cat box selectionMobileSAM · box Horse point selectionMobileSAM · multi-subject scene
About this showcase: These are real API outputs presented as visual demonstrations, not model-quality benchmarks. Reproducible API tests, verification reports, and integrity evidence are maintained separately in the project's public verification repository.
Verification & reproducibility → Public verification repository