Quick, Visual or AI Agent: Which Segmentation Mode Should You Use?
9/21/2026
The online segmentation workspace now has three modes: Quick Mode, Visual and the new AI Agent. Each one suits a different kind of image and request. This guide explains what each mode does, what it costs and when to switch.
The three modes at a glance
| Quick Mode | Visual | AI Agent (new) | |
|---|---|---|---|
| How you choose the target | Pick from the objects found in your image | Click points, draw a box or a loop | Describe it in a sentence |
| Best for | Clear, common objects | Exact parts, objects without a name, crowded scenes | Specific requests such as "the person on the left, including their bag" |
| Speed | Fastest | Fast | Slower: it works in several steps |
| Follow-up edits | Pick another object | Add include or exclude points | Keep chatting: refine, hide, undo |
| Credits | 1 to identify the objects in a new image, then 1 per object | 1 per Visual Segment | Per message: conversation usage, rounded up, plus 1 per segmentation |
All three modes live on the same Online Segment page. Switch between them with the tabs above the canvas.
Quick Mode: pick from what's in the image
Upload an image and SegmentLens identifies the objects in it automatically. They appear as a list of names, such as cup, laptop or sneaker. Search the list, click a name, and that object is cut out.
- No typing. You choose from what is actually in the image.
- Only objects that can be cut out are listed. Names the model cannot segment are left out, so a click should not come back empty.
- Picking the same object again on the same image is free.
Use it for product photos, portraits and any image where the thing you want is an obvious, nameable object.
Visual: point at exactly what you want
Visual mode lets you mark the target on the image yourself:
- Include points (+) mark areas that belong to the object.
- Exclude points (−) mark nearby areas to ignore.
- Boxes and closed loops enclose a target. Draw separate loops for separate targets, up to 15 at a time.
Once an object is segmented, you can refine it with more clicks on missing or unwanted areas. Visual is the right choice when the object has no simple name, when you only want part of it, or when two similar objects sit side by side. See Click-to-Segment for a detailed walkthrough.
New: AI Agent, segmentation by conversation
AI Agent mode lets you describe what you want to keep, in plain words, and keep working on the result in a conversation.
- Switch to the AI Agent tab and upload an image.
- Describe what to keep, for example "Keep the red car on the left" or "Keep the person on the left, including their bag".
- Send. The agent studies the image, segments the target and checks the result against your request. When a first attempt misses, it rewrites the request and tries again on its own.
On hard images, this is where it beats Quick Mode: it can handle position, color, size and "with their…" details that a single object name cannot express.
Keep editing in the same conversation
- Refine a result. Choose an existing result and describe the correction, such as "include the handle too".
- Click an object to focus it on the canvas, × to hide it, and Undo to restore the last edit.
- Every result keeps its versions. Preview an earlier version and continue from it at any time.
- Download the objects you selected as transparent PNGs.
Pick up where you left off
Each conversation is saved with its image, messages and result versions. Resume your last conversation, or open any earlier one from the conversation history. Changing the image starts a new conversation, and your previous ones stay in history.
What it costs
- The conversation is charged by actual usage and rounded up to whole credits for each message, with a minimum of 1 credit when a charge applies.
- Each segmentation the agent runs costs the usual 1 credit. One message runs at most two.
- Example prompts, focusing, hiding and Undo are free.
The actual charge is shown after every message. You can stop after the current step at any time, and retrying a request that was interrupted does not charge you twice.
Which mode should I use?
- The object you want appears in the list after upload. Use Quick Mode. It is the fastest and cheapest way.
- You need an exact part, an object without a name, or a precise boundary. Use Visual.
- You can describe the target but it is not in the list, it needs a qualifier such as "left", "largest" or "red", or you expect several rounds of changes. Use AI Agent.
- Quick Mode or Visual came back empty. Click Try AI Agent. Your current image comes with you, along with a draft request, so there is nothing to upload again.
You can also go the other way: inside AI Agent, Open visual tools takes you to point-and-click marking, and Exit Agent returns you to the other modes.
Tips for better AI Agent results
- Say what to keep, not how to do it. "Keep all the people" works better than step-by-step instructions.
- Add one distinguishing detail when there are similar objects: position, color or size.
- Correct instead of starting over. "Only the left one" or "also include the shadow" builds on the current result.
- Know the limits. Text, labels and some medical terms often return nothing. For those, mark the region in Visual mode instead.
Get started
- Open the segmentation workspace and choose a mode tab
- Read the full changelog for version 1.11.0