Client Side Forensic Utility With Zero Data Upload
Free Video Analyzer
Find out quickly whether a clip deserves a closer look. The footage stays on your device.
Start a video check
The players, the sampled frames, the optional local AI frame experiment, the executive triage and every collected record appear here once you choose a clip or load the sample. Your video stays on this device.
The analyzer reads timing patterns and sampled visual changes, compares the clip with a reference when you have one, optionally runs a local image model on selected frames and states plainly what was tested and what remains unknown.
Choose the clip, and a reference when you have one
Select the clip in question and, when one exists, the earlier or original copy to compare against. The analyzer hashes the supplied bytes, reads the container and track records, examines every packet timestamp in the selected range, decodes a bounded set of sample frames and summarizes the observations in an executive triage. Optional browser local image classifiers appear after the results; you choose one by its task and download size, and nothing is sent to an AI provider.
Zero Upload Local Browser Processing
Supported containersMP4, M4V, MOV, WebM, MKV, TS, MTS and M2TS, plus Dahua DAV and raw H.264 or H.265 streams decoded locally by GDF readers
Size limit2 GiB per video, 60 seconds and up to 120 sampled frames per range
EngineMediabunny container reader with browser WebCodecs decoding
Container recognition reads the bytes, not the extension. Whether frames can be decoded depends on the codec and the browser you are using: HEVC support varies by device and no browser decodes ProRes, in which case the analyzer still returns the container records and the SHA 256 and marks frame work unavailable.
Read the observations beside their limits
Clips copied straight off a recorder open here too: Dahua DAV exports and raw H.264 and H.265 streams are decoded on your device by readers GDF wrote, and the report names the reader beside every other record. A DAV file adds the recorder date and time from each frame header, with any jump or reversal listed, while a raw stream carries no timestamps, so its timing is reported as frame order only.
Video Analyzer opens in Guided mode: the plain findings, the executive triage and the observation cards come first, with five frame views, the optional image classifiers and the Try it with AI panel. Switch the same workbench to Examiner mode and all fifteen frame views, the complete record browser and the full reference comparison detail appear without checking the clip again; the page remembers your choice in this browser and the report records the mode you used.
Every flagged item carries a Show me button that moves the player to that exact moment, decodes the frame there and opens the view that shows it best, so you can look at a gap or a cut yourself rather than take the table on trust. The sample behind Load Sample Forensic File is a training clip with one planted edit, three seconds removed at the nine second mark, and the results say so.
Start with the supplied file hash, container records and codec information in the executive triage. The analysis cards then show timing irregularities, repeated and near identical samples and substantial scene or color changes; samples can be seconds apart, so a repeated or near identical sample is not a test for dropped or duplicated frames. Each card opens to the underlying timestamps and settings. A repeated frame can come from a static scene, a scene change can be an ordinary cut, movement or flash, and a color change can follow lighting, exposure, processing or a filter. These observations point to material you can inspect; they do not identify a proven edit.
No reference is required
A single video check records the facts available in that file and its sampled decoded frames without manufacturing an earlier version. Regular timing cannot exclude deleted frames or a speed change when an editor has rewritten timestamps. With a reference, the tool adds ordered candidate frame matches, unmatched sampled runs and timing ratios between aligned content. Read the ambiguity warnings before treating a match as useful, because the tool has not authenticated the reference either.
Compare content and presentation timing
For reference comparison, choose corresponding ranges in both files. Matching uses reduced 32 by 32 pixel descriptors rather than every original pixel, so different framing, inserted material, overlays, reencoding and visual processing can change correspondence. The reported relative presentation rate describes matched sample anchors; it can point to a speed difference, but sparse samples and repeated content can make the estimate ambiguous, and unmatched samples alone do not establish additions or removals.
Inspect a selected frame
Use the sampled frame timeline to open a full supported decoded frame and inspect channels, contrast, edges or residuals with the split screen slider. Difference views require equal dimensions and use the previous sample or a reference sample at the same relative range progress; they do not automatically register moving objects. The views operate on browser decoded 8 bit display pixels, do not preserve original HDR precision and do not provide sensor PRNU. JPEG error level analysis and stored PNG hidden color analysis are not treated as video frame authenticity tests.
Choose an optional local model
The small generated image model downloads about 10.1 MiB of weights. CIFAKE, face image and three label alternatives each download about 54.1 MiB, and shared runtime code adds about 13.6 MiB on first use. A larger model is not automatically more accurate. Select one model, read its training task, confirm the limits and choose up to 12 frames; only Run selected model locally fetches those weights. Processing uses one WASM CPU thread, and the face model applies the same confirmed square rectangle to every selected frame, so check that the face stays inside it.
AI results describe individual frame experiments
These four models classify still images and have not been evaluated by GDF as temporal video detectors. A brief edit can be missed between selected frames and an ordinary image can receive a misleading label. No API key or hosted AI service is used. Each frame retains its raw uncalibrated labels and logits, input and model hashes, crop, preprocessing and runtime information, with errors and cancellations recorded separately. There is no probability of being fake, calibrated confidence or combined authenticity score.
Try it with AI, offered in beta under the results, runs a small language model on your own graphics processor after you confirm a one time download from this website of about half a gigabyte. It receives only the numbered observations and the methods text of your inspection, restates them in plain English without adding a fact and answers questions about them. Its sentences appear in a labelled box as generated text, are excluded from the findings table, the triage and the PDF, HTML and CSV reports, and are never a finding about authenticity. Current Chrome or Edge on a desktop with WebGPU is required.
Save the complete record
Download the PDF or HTML report in the Global Digital Forensics four section layout: executive triage and file identity with the full SHA 256, the plain English interpretation guide, the structured findings inventory with your notes, and the evidentiary scope advisory. The Excel ready CSV gives one clean column per field and the JSON is the lossless record of source hashes, track timing, sampling coverage, frame timestamps, applied views, reference correspondence, notes and AI attempts. A display search never removes records from an export. The PDF and HTML reports carry reduced copies of sampled frames and the applied view as figures, the JSON and CSV files hold text records, and no report contains the footage itself; save a view PNG separately when you need the full size picture.
No analysis upload or marketing transmission
The selected videos, file names, hashes, decoded frames, notes and report data never reach an analysis server. Optional usage measurement stays off unless you allow Analytics. It sends GDF only the tool name and fixed action labels, such as analysis completed or report prepared. No input, result, filename or visitor identifier is sent. Google and advertising scripts do not load on this page. Use Cookie settings below to change your choice. Program files, fonts and the optional model weights you choose download from this website without attaching your footage.
Clear closes the local inspection and removes displayed results while previously saved files remain on your device. Review private details before sharing a report. Local processing protects the input from an analysis upload; chain of custody comes from how the original is preserved.
Browser, format and sampling requirements
Video formats and codec support
MP4 and M4V use the same container family, and MOV is a QuickTime container whose video may be H.264, HEVC, ProRes or another codec, so the extension alone never establishes that this browser can decode it. MKV and TS, MTS or M2TS clips can support local frame inspection even when the built in player refuses to play them, because playback and analysis have separate browser requirements. Transport streams carry no descriptive tags through this reader; that is unavailable coverage, not evidence that tags never existed.
No browser decodes ProRes. HEVC and other codecs depend on your browser and device, and AVI, WMV, FLV and OGV video sit outside the current reader. Renaming or converting a file does not restore original recording information; preserve the original, because a converted working copy is a different file with a different SHA 256.
Accepted containers are MP4, M4V, MOV, WebM, MKV and the MPEG transport stream family, recognized from the bytes of the file. Whether frames can be analyzed depends on the codec, its profile and the browser in use. A track the browser cannot decode still yields its container records and the SHA 256 of the file, with frame work marked unavailable, and a current desktop Chromium browser offers the broadest tested path.
Each file is limited to 2 GiB and four million coded and displayed pixels, and it is read and hashed in 8 MiB parts so a large export never has to fit in memory. Each selected range is at most 60 seconds with 2 to 120 sampled frames and at most 12,000 encoded packet timing records. The operation time limit is 120 seconds, with up to 128 more seconds for each GiB while the file is hashed, and optional AI is limited separately to 12 frames and five minutes per batch.
Coverage is recorded, not assumed
Descriptors and contact sheet previews use explicitly recorded reduced dimensions, while selected inspection and AI frames use full supported decoded display pixels. Decoding and browser color conversion can alter the pixel representation, and HDR, original high bit depth values and all frame continuity are outside this tool. PDF export stops above 25,000 records, 500 pages, 25 MiB or 60 seconds with an explanation rather than a shortened PDF.
Edit detection accuracy on real world footage has not been measured by GDF, so an unavailable or cancelled result is unknown rather than a finding that nothing was edited. Tell us about a problem with the tool; the link attaches no selected file or report. Video Forensics offers the hands on workbench, Photo Analyzer covers still images, Metadata Analyzer reads the properties of any recognized file, and local container parsing uses Mediabunny with decoding described in WebCodecs. Version 1.2.0. All free tools. © 2026 Global Digital Forensics, Inc. All rights reserved.
Free consultation
Does the clip need an examiner’s eyes
When the triage flags a single long gap in the timing, unmatched segments or a speed ratio that matters to your matter, a senior examiner can tell you on the first call what the observations mean, what to preserve and how to transfer the originals securely. Our examiners have testified in federal and state courts on video evidence. Your selected video and report are never attached.
Reviewed by Joseph Caruso, Founder, CEO and CTO, CISSP. Last reviewed .