Follows objects through a video
Detects, marks and tracks objects from frame to frame.
Find objects and outlines
Sam 3 · Video→Video
SAM 3 is a unified foundation model for promptable segmentation in images and videos. It can detect, segment, and track objects using text or visual prompts such as points, boxes, and masks.
Made by Meta50 creditsper run Pro
Opens the video studio with Sam 3 · Video→Video ready to use. You can look around as a guest. Running it takes the Pro plan or higher.
What Sam 3 · Video→Video makes, what it needs and what you get
What it makes
Find objects and outlines
Finds objects, depth, poses and outlines in a video.
What you need
- The video to work on
Optional: what to find and the area to change.
What you get
Masks, depth maps or outlines
About Sam 3 · Video→Video
Sam 3 finds objects in a video, marks them and follows them from frame to frame. Marking which parts of each frame belong to an object like this is called segmentation. You can point it at things with a text prompt, points or boxes. The result is the segmented video.
Sam 3 has two modes: Video to Video and Video to RLE. Both need one video. A Prompt names what to track, and commas separate several objects, for example person, cloth. Point Prompts and Box Prompts mark objects by position. Detection Threshold sets how sure a detection must be, and lower values find more objects but are less precise. Video to Video lets you pick MP4 or WebM output. Video to RLE adds a Mask image, a Frame Index and a Boundingbox Zip option. Frame Index sets the frame used first when you add a mask. Boundingbox Zip returns a zip file with a bounding box (a rectangle around an object) overlay for each frame.
Sam 3 costs 50 credits and takes one video at a time. Sam 3 1 Video to RLE costs 100 credits and adds a Max Num Objects option.
Other models for this job
Same job: find objects and outlines. Compare before you choose.
Video Pro Open
Sam 3 1 Video to RLEFind objects and outlines
SAM 3.1 builds comes with Object Multiplex, a shared-memory approach for joint multi-object tracking that delivers faster speeds with larger number of objects tracked.
Meta100 creditsper run
Video Pro Open
Depth Anything VideoFind objects and outlines
Generates depth maps from video using Video Depth Anything (CVPR 2025). Produces per-frame depth estimation with temporal consistency across frames. Supports 3 model sizes (Small, Base, Large), 5 colormaps including grayscale, side-by-side comparison with the original video, and raw depth export as .npz. Useful for 3D reconstruction, video effects, compositing, and scene understanding.
Open Source2,000 creditsper 5 s
Video Pro Open
DWPose Pose Prediction Video toFind objects and outlines
Predict poses from videos.
DWPose6 creditsper run
What it does well
Several objects at once
Separate objects with commas in the prompt, for example person, cloth.
Text, points, boxes or a mask
Point Prompts and Box Prompts mark objects by position, and Video to RLE also accepts a Mask image.
How to use Sam 3
How It Works
How It Works
01
Add your video
Upload the one video you want to segment.
02
Say what to follow
Choose Video to Video or Video to RLE. Type a Prompt, or add points or boxes.
03
Generate
You get back the segmented video. It costs 50 credits.
Tip In Video to Video, Video Output Type lets you choose MP4 or WebM.
Good for
Pick out people or objects
Name what to follow in the prompt, for example person, cloth, and have it marked in the video.
Point at what to follow
Add Point Prompts or Box Prompts to mark where an object is, and have it followed through the video.
Questions
Quick answers to help you get the most out of Sam 3 · Video→Video in FunAI Studio.
1. What do I need to provide?
2. What do I get back?
3. How much does it cost?