What is an image segmentation map?

It is a version of your photo in which every region the model finds is filled with one flat colour. It describes where things are and where each one ends, and discards what they looked like. Generation models read it as structural guidance.

Can I use this to cut one object out of a photo?

No, and this is the difference worth knowing before you spend a credit. This tool segments the entire frame automatically and returns one colour-coded map. It cannot isolate a single named object, and the regions are not exported as separate masks. If you want one subject lifted out, use a cutout or background removal tool instead.

Does it tell me what each region is?

No. SAM finds boundaries rather than categories, so a region is marked as distinct without ever being named. There is no list of detected classes and no machine-readable export — if you need labelled categories, this is not the right tool.

Can I use the output as a ControlNet seg input?

Yes, that is what most people generate one for. The map is a standard colour-coded segmentation image, so it works anywhere a seg conditioning input is expected. Because the result opens in a generation session, you can also wire it into a model here without downloading it first.

How many regions will it find?

It depends entirely on the photo, and the number is not something you set. A busy street produces many regions; a subject on a plain backdrop produces very few. Pictures with several clearly distinct areas give the most useful maps.

What image formats can I upload?

JPG, PNG and WebP, one image per run. The segmentation map comes back as a standard image you can download or pass straight to another node.

Is my photo processed in my browser?

No. Some segmentation tools run entirely on your device; this one does not. Your photo is uploaded and segmented on our servers, which is what allows the same SAM model to run regardless of the machine you are on. If keeping the file off a server matters more to you than model quality, a local tool is the better choice.

AI image segmentation

AI Image Segmentation: Turn a Photo into a Segmentation Map

Upload a picture and SAM segments the whole frame, painting every region it finds a different flat colour. You get a segmentation map you can feed straight into a generation model as structural guidance — no prompt to write, nothing to click.

2 credits

Street photo with buildings, road and parked vehicles

Segmentation map of the same street, each building, the road and each vehicle a separate flat colour

Examples

Segmentation Map Examples

Street scene

How it works

Create a Segmentation Map in Three Steps

Step 1
Upload a photo
Drop in one JPG, PNG or WebP. Pictures with several distinct areas — a street, a room, a figure against a background — give the clearest maps.
Step 2
Let SAM segment it
The whole frame is segmented automatically. There is no prompt to write and no object to select, so the same photo gives the same map every time.
Step 3
Download or keep working
The map opens in a generation session, so you can download it or wire it straight into a model as structural guidance without starting again.

What you get back

How to Read a Segmentation Map

Looking to cut one named object out of a picture instead? That is a different tool — the colours below are regions, not objects you can lift out individually.

Regions, not labels

Each colour marks an area the model considers one thing. The colours are arbitrary and carry no names: nothing here tells you that a region is a car or a tree, only that it is separate from what surrounds it.

The whole frame, automatically

Every region is found in one pass. You do not click points, drag boxes or type what to look for — which is the difference between this and most tools that share the phrase image segmentation.

Boundaries, flattened

Colour, texture and lighting are discarded and only the layout survives. That loss is the point: a generation model reading this map inherits the structure of your photo without inheriting how it looked.

What it is for

What People Use a Segmentation Map For

ControlNet seg conditioning

The most common use by far. A seg map hands a generation model the layout of a real photo, so a new image keeps the arrangement while everything about its appearance changes.

Planning a composite

A flat map of where each area begins and ends is easier to reason about than the photo itself when you are working out what sits in front of what.

Seeing how a model reads a scene

The map shows which parts of the frame a segmentation model treats as one thing. Where it merges or splits regions unexpectedly tells you something the photo does not.

Blocking out colour and shape

Stripped of texture and lighting, a photo becomes flat masses of colour — a quick way to study composition or start a design from a real arrangement.

Credits and plans

Each segmentation map costs a flat 2 credits, whatever the size of the photo.

Basic

For hobbyists and explorers

$9.99/mo

Select Plan

1000 credits
Up to ~200 images/month
At least ~1000s video/month
Intelligent creative agent
Video editor
Latest image models, including GPT Image 2.5, GPT Image 2 and Nano Banana Pro
Latest video models, including Seedance 2
Realistic face uploads
Voice generation with ElevenLabs
No watermark
Unlimited template access

Pro

Max

For power users and teams

$149.99/mo

Select Plan

18000 credits
Up to ~6000 images/month
At least ~18000s video/month
Intelligent creative agent
Video editor
Latest image models, including GPT Image 2.5, GPT Image 2 and Nano Banana Pro
Latest video models, including Seedance 2
Realistic face uploads
Voice generation with ElevenLabs
No watermark
Unlimited template access

Team

For teams and studios

$329.99/mo

Select Plan

36300 credits
Up to ~12000 images/month
At least ~36300s video/month
Realistic face uploads
Share canvas, workflows, and assets with your team
Up to 10 members per team
Up to 5 teams
Role-based management
Team credit management and spending caps
Per-member usage tracking

See all plans

FAQ

Image Segmentation Questions

What is an image segmentation map?: It is a version of your photo in which every region the model finds is filled with one flat colour. It describes where things are and where each one ends, and discards what they looked like. Generation models read it as structural guidance.
Can I use this to cut one object out of a photo?: No, and this is the difference worth knowing before you spend a credit. This tool segments the entire frame automatically and returns one colour-coded map. It cannot isolate a single named object, and the regions are not exported as separate masks. If you want one subject lifted out, use a cutout or background removal tool instead.
Does it tell me what each region is?: No. SAM finds boundaries rather than categories, so a region is marked as distinct without ever being named. There is no list of detected classes and no machine-readable export — if you need labelled categories, this is not the right tool.
Can I use the output as a ControlNet seg input?: Yes, that is what most people generate one for. The map is a standard colour-coded segmentation image, so it works anywhere a seg conditioning input is expected. Because the result opens in a generation session, you can also wire it into a model here without downloading it first.
How many regions will it find?: It depends entirely on the photo, and the number is not something you set. A busy street produces many regions; a subject on a plain backdrop produces very few. Pictures with several clearly distinct areas give the most useful maps.
What image formats can I upload?: JPG, PNG and WebP, one image per run. The segmentation map comes back as a standard image you can download or pass straight to another node.
Is my photo processed in my browser?: No. Some segmentation tools run entirely on your device; this one does not. Your photo is uploaded and segmented on our servers, which is what allows the same SAM model to run regardless of the machine you are on. If keeping the file off a server matters more to you than model quality, a local tool is the better choice.