How do I get the most accurate masks from a text prompt?

Be specific and descriptive. Include distinguishing attributes like color, clothing, or object type (e.g., "red sedan" instead of "car"). If multiple similar objects appear, refine the prompt further ("person in blue jacket" vs. "person").

Can it track multiple instances and keep IDs stable?

Yes. The SAM3 Video Masking node finds all instances that match the prompt and tracks them with stable IDs across frames. If you only need one instance, make the prompt more specific or filter by ID in post using the raw mask stream.

It works on a few frames but is slow on the full video. Any tips?

Use frame_load_cap for quick iteration, then switch to 0 for final output. Reducing input resolution or trimming the clip can improve speed. Running on a GPU with sufficient VRAM is recommended for real-time responsiveness.

What should I export for compositing in another editor?

Export the raw mask (as a video matte) and the cut-out subject. In your NLE or VFX app, use the mask as a luma/alpha matte, or place the cut-out directly over your background plate for precise compositing.

SAM3 - Video Masking via Text Prompt

This ComfyUI workflow performs automatic video object masking from a simple text phrase using the SAM3 Video Masking node (ID: 9cea40bb-b0cf-4b40-a758-8935cfe8d52f). You load a clip with VHS_LoadVideo, describe what you want to isolate (for example, "red car" or "person") in the node’s mask_prompt, and the system finds every matching instance, tracks them across frames with stable IDs, and produces three streams: a mask overlay preview, a raw per-frame binary mask, and a subject cut-out. SaveVideo nodes handle exporting the results.

Technically, VHS_LoadVideo provides a consistent frame sequence (fps, resolution) for inference. During iteration you can cap frames using frame_load_cap to preview quickly, then switch to a full pass by setting it to 0. The SAM3 Video Masking node interprets your text prompt, segments the target across frames, and maintains identity continuity so objects don’t flicker or swap. The outputs are suitable for downstream compositing, selective blurring, color grading, or background replacement. A MarkdownNote node in the graph provides inline tips and reminders while you work.

API

Use this workflow from code

Every Comfy workflow is a JSON graph. The payload below is this workflow, exactly as ComfyUI runs it — fetch it from the URL, keep it in version control, or load it in ComfyUI and run it node by node.

4d7abcb8e25d.json

Fetching workflow JSON…

Run it from TypeScript or Python with the Comfy SDK. The same code targets Comfy Cloud or a ComfyUI you host yourself — only the base URL changes.

// Install (beta)
npm i @comfyorg/sdk

// Run this workflow (TypeScript)
import { Comfy } from "@comfyorg/sdk";

const client = new Comfy({ apiKey: "comfyui-..." });
const wf = await client.workflows.fromFile("workflow_api.json");
const job = await client.run(wf);
await job.getOutputs("<output-node-id>")[0].toFile("output.png");

The SDK takes a workflow in API format: open this workflow in ComfyUI and use File → Export Workflow (API).

Comfy Cloud API access requires a plan with an API key.

SDK docs Get an API key

FAQ

Frequently Asked Questions