If you are a researcher, engineer, or QA professional, you know the frustration of loading heavy, resource-intensive software just to compare two video clips. Whether you are evaluating a new compression algorithm, auditing rendering output, or benchmarking an AI video upscaling model, you need a workflow that is fast, accurate, and lightweight.
TwinLens is a browser-based comparison tool that runs entirely locally — no uploads, no servers, no sign-in. Drop in 2, 3, or up to 4 videos and you are ready to analyze within seconds.
Getting started: drop your videos
Navigate to the TwinLens comparison page and drop your video files into the panels — up to four at a time. TwinLens automatically synchronizes playback, normalizes display size across all clips, and keeps them aligned with continuous drift correction. The toolbar switches into video mode and every clip begins looping in sync immediately.
For 3 or 4 clips, TwinLens lays them out in a horizontal row, vertical stack, or 2×2 grid — perfect for ablation studies, multi-checkpoint comparisons, or A/B/C/D variant reviews.
Views, metadata, and zoom
TwinLens gives you three comparison views — side-by-side, slider, and peek — plus per-clip metadata (i for codec, resolution, frame rate, duration), editable labels (T) that carry into exports, and a size toggle between fit-to-viewport and actual 1:1 pixels for evaluating sharpness at native resolution.
Exporting your comparison
Once you have tuned your view, TwinLens gives you two export paths:
- Snapshot — pauses playback and saves a PNG of the current comparison frame, including any active labels and view mode. Use this to capture a specific artifact or frame of interest.
- Record — starts and stops a screen recording of the active comparison view, saved as a video file. You can adjust playback speed before recording, so a slow or fast source clip can be captured at a viewer-friendly rate. The output is ready to drop directly into a slide deck, Notion doc, or bug report.
Typical workflow for ML video evaluation
A common pattern when comparing model outputs:
- Drop the reference clip (ground truth or baseline) on the left and the model output on the right.
- Use actual size mode and switch to slider view to scan for sharpness differences and temporal artifacts frame by frame.
- Add labels via the T icon — e.g. baseline vs SeedVR2 4×.
- Snapshot key frames that illustrate regressions or improvements.
- Record a short looped segment at 0.5× speed for your model card or PR review.
TwinLens is built to make technical image and video comparison faster and more intuitive — entirely in the browser, with no data leaving your machine. If it improves your workflow, consider recommending it to your team.
Related tutorials
Ready to compare your videos?
Free, private, and instant — drop in up to 4 video files and start analyzing.
Open TwinLens — No Install Needed →