InVID verification is a practical workflow for turning a video into searchable evidence. The InVID-WeVerify/VeraAI browser extension can extract representative keyframes from a video, send those frames to reverse image search engines, expose useful context and metadata, and help investigators trace footage back to earlier versions or a more credible source.
The tool is especially useful when a viral clip may be old, cropped, reposted with a false caption, taken from a longer video, or detached from its original event. It does not automatically tell you whether a claim is true. Its value is that it converts a difficult “Where did this video come from?” problem into a set of smaller searches that a human can verify.
Open InVID-WeVerify/VeraAI, use the Keyframes tool with a supported video URL or local video file, then reverse search several representative frames. Prioritize frames with signs, logos, landmarks, people in context, distinctive objects, or scene changes. If the first frames fail, request more keyframes. Treat every search result as a lead, compare publication dates and longer versions, and distinguish an earlier known copy from the true original source.
What Is InVID Verification?
The tool commonly called InVID or InVID-WeVerify is a browser-based verification toolkit originally developed for journalists and fact-checkers. The current Chrome extension is listed as Fake news debunker InVID WeVerify VeraAI, reflecting its development across the InVID, WeVerify and vera.ai projects.
The extension is broader than a keyframe extractor. Depending on the current build and available services, its toolset includes video and image context analysis, keyframe extraction, reverse image search, metadata inspection, image forensic functions, OCR and other verification utilities. The current extension listing is available in the official Chrome Web Store.
For video verification, however, the most important feature is simple: fragment a video into useful frames and search those frames independently.
Why InVID Is Useful for Viral Video Verification
Search engines are much better at matching still images than accepting a random social video and reconstructing its entire history.
InVID bridges that gap.
| Problem | What InVID helps you do | What still requires human verification |
|---|---|---|
| You have a video but no source | Extract representative frames for search | Determine which result is closest to the original source |
| The clip may be old | Find matching frames on older pages | Interpret whether those dates predate the current claim |
| The clip is cropped | Search frames that still contain distinctive scene information | Compare versions and identify what was removed |
| The viral caption may be false | Connect the visual to other contexts online | Verify the event, date, location and source independently |
| The first search returns nothing | Generate or inspect additional keyframes | Choose better frames and change the search strategy |
What Are InVID Keyframes?
InVID’s keyframe workflow extracts representative still images from the video. The official InVID documentation describes the number and choice of keyframes as depending on variation in the video’s visual content.
That matters because a useful keyframe set is not the same as taking a screenshot every five seconds.
A 30-second clip may spend 20 seconds on one speaker, then briefly show a sign, crowd, vehicle, building, logo or wider scene. A good fragmentation workflow gives you a better chance of capturing those visually distinct moments.
The official InVID Verification Plugin documentation explains that the Keyframes service can process a supported video URL or an uploaded file and provides additional keyframes on demand.
Keyframe Does Not Mean “Proof”
A keyframe is a search input.
It does not prove:
- where the video was recorded
- when the scene occurred
- who first uploaded it
- whether the caption is accurate
- whether the content was generated by AI
A successful keyframe search creates a connection between your video and other indexed media. The next step is to evaluate that connection.
Which Keyframes Are Best for Reverse Search?
Choose frames with high information density.
If you want a tool-neutral explanation of frame selection itself, use the video frame extractor guide. This page stays focused on the InVID workflow.
How to Use InVID to Verify a Viral Video
-
Preserve the claim before searching.
Record the post URL, account name, visible publication date, caption, claimed location, claimed event and any watermark. Your investigation should test that claim rather than silently replacing it with whatever the search engine returns.
-
Open the InVID-WeVerify/VeraAI extension.
Use the current official extension rather than an unrelated tool using a similar name. The interface evolves, so focus on the Keyframes function rather than relying on the exact position of a button.
-
Open the Keyframes tool.
This is the core workspace for video-source tracing. AFP’s own verification tutorial uses the InVID-WeVerify Keyframes section as the starting point for reverse video search.
-
Submit a video URL or local file.
For a public video, paste a supported URL when the service can access it. If the URL cannot be processed, use a local copy that you are authorized to handle. Platform restrictions, login requirements and expired posts can affect URL processing.
-
Wait for the representative frames.
The service may take time depending on the source and video. Do not assume an empty first screen means the video has no searchable evidence.
-
Scan all frames before clicking search.
Choose visually different frames. One may identify a person, another a location, and another an event or publisher.
-
Reverse search the strongest frames.
The plugin can route keyframes into supported reverse image search services. Search several frames, because each frame produces a different visual query and therefore a different result set.
-
Request more keyframes when necessary.
If the initial set is visually repetitive or misses a critical scene, use the additional-keyframes option where available. AFP specifically recommends trying more keyframes rather than giving up after the first poor result.
-
Open the matching pages.
Do not decide from thumbnails. Inspect the page, publication date, video length, description, source credit, watermark and whether the same scene is actually present.
-
Trace backward from the best match.
If a result credits another account or source, follow that attribution. Your goal is not the first page Google shows. It is the strongest reconstructable source history.
AFP’s practical guide demonstrates this exact logic: extract keyframes with InVID-WeVerify, reverse search different frames, request more keyframes when needed, and then investigate the resulting pages rather than treating the tool as an automatic verdict. See AFP’s guide to finding the source of a video.
URL or File: Which Input Should You Use?
| Input | Use it when | Limitation |
|---|---|---|
| Public video URL | The video is still accessible and the service can retrieve it | Platform restrictions, authentication or URL changes can block processing |
| Local video file | You have the actual clip or a preserved download | The local copy may already be cropped, transcoded or stripped of source context |
When possible, keep both the original post context and the best available media copy. A local file is useful for frame extraction, while the original URL is useful for source and chronology.
Why You Should Search More Than One Keyframe
Each keyframe represents a different query.
Imagine a 45-second viral video:
- Frame A shows only smoke.
- Frame B shows a distinctive tower.
- Frame C contains a television watermark.
- Frame D shows an emergency vehicle with readable markings.
Frame A may return generic disaster imagery. Frame C may identify the broadcaster immediately. Frame D may reveal the city. Searching only the first frame can therefore create a false impression that the video is untraceable.
What “More Keyframes” Is For
InVID’s additional-keyframe option is useful when the initial fragmentation:
- contains too many visually similar frames
- misses a short but distinctive scene
- captures transition frames instead of useful context
- does not produce searchable results
More keyframes are not automatically better. The goal is to discover different evidence, not create dozens of near-duplicates.
Reverse Search Strategy: Search Broad, Then Search Specific
Start with a frame that preserves the whole scene. If results are weak, isolate a stronger visual clue using the search engine’s own crop or region-selection tools.
A practical sequence is:
- search the whole representative frame
- open exact or close matches
- crop around a logo, sign, landmark or object if necessary
- search another keyframe from a different scene
- combine visual results with text from captions or signs
For a deeper workflow specifically around Google’s visual search interface, use the Google Lens video search guide.
How to Read InVID Reverse Search Results
The quality of the investigation depends on how you interpret the results.
| Result | What it supports | What it does not prove |
|---|---|---|
| Same frame on an older page | The visual existed online by that earlier date | The exact recording date or original uploader |
| Longer version of the same clip | Your viral version was excerpted or cropped from a longer source | That the longer upload is necessarily the first upload |
| Same footage with another location claim | At least one context claim deserves investigation | Which location is correct without geolocation evidence |
| Matching news article | A credible lead for event, date or source context | That every detail in the article is automatically correct |
| No useful match | The searched frame did not produce a useful indexed result | That the video is new, original, authentic or AI-generated |
The Earliest Result You Find Is an Earlier Known Occurrence
This wording matters.
If InVID leads you to a video uploaded in 2021 and your viral post claims the footage was recorded this week, you have strong evidence against the “new footage” claim.
But the 2021 upload may itself be a repost.
Write:
“The same footage was publicly available by 2021.”
Do not automatically write:
“The video was filmed in 2021.”
If reconstructing origin, transformations and versions is the main task, continue with the video provenance guide.
Use Keyframes to Find the Longer Context
One of the most valuable InVID outcomes is discovering that a viral clip is only a short excerpt.
A longer source can reveal:
- what happened before the viral moment
- what happened after it
- who recorded or published it
- the actual event or location
- whether a scene was staged, fictional or part of a performance
AFP has repeatedly used InVID keyframes to find longer or earlier versions of misleading clips, including cases where a scene was actually from a film, animation or old event rather than the story attached to the repost.
Keyframes Can Expose Old Footage Reused as Breaking News
This is one of the highest-value use cases.
Suppose a post says:
“Breaking: this is happening right now.”
If a reverse search of an InVID keyframe finds the same scene in a credible article or upload from years earlier, the current-event claim fails even if the pixels in the video are genuine.
This is why source tracing often solves misinformation faster than deepfake analysis.
Use Text and Watermarks After the Visual Match
A keyframe can also reveal searchable text:
- broadcast logos
- usernames
- street signs
- subtitles
- lower-third graphics
- event names
- product or company branding
Once you identify a clue visually, search its exact wording outside InVID too. A watermark may lead directly to the source account even when reverse image search results are weak.
InVID Metadata Is Supporting Evidence, Not a Truth Certificate
The toolkit can expose metadata and contextual information in supported workflows, but metadata should be interpreted according to where it came from.
A downloaded or platform-processed video may contain dates and encoder information describing the current copy rather than the original camera recording.
Use metadata to generate questions such as:
- Was this copy re-encoded?
- Does the date conflict with the claim?
- Is the current file structurally different from a presumed original?
Do not treat one timestamp or software tag as an authenticity verdict.
What About the Forensic Filters?
InVID includes image-oriented forensic tools designed to help inspect still images and extracted frames.
These can be useful for exploring suspicious regions, but they do not replace a complete video-forensics examination. A visual anomaly may come from compression, resizing, screenshotting, social-media processing or ordinary image enhancement.
The correct role of a forensic filter is:
“This region deserves closer investigation.”
not:
“This filter proves the video is fake.”
InVID Is Not a Deepfake Detector
InVID can help find where footage appeared before, reveal context and provide still-image analysis tools. That is different from model-based AI video detection.
A real video can be reused with a false caption. A synthetic video can be so new that no earlier search match exists. A manipulated video can contain mostly authentic footage.
When the unresolved question is whether the media itself contains AI-generation or manipulation signals, DetectVideo AI can contribute a separate technical analysis layer.
InVID vs Video Frame Extractor vs Reverse Video Search
| Tool or method | Primary job |
|---|---|
| InVID verification | Tool-specific workspace for extracting keyframes, searching frames and inspecting verification clues |
| Video frame extractor | General method for turning a video into searchable stills |
| Reverse video search | Broader source-tracing process using frames, text, audio, platform search, chronology and source history |
This separation is important. InVID is one powerful tool inside a wider source-verification process. The reverse video search guide owns that broader workflow.
What If the Video URL Does Not Work in InVID?
Do not assume the tool is broken or the video is suspicious.
A URL may fail because:
- the post requires login
- the platform changed its delivery method
- the video is private, deleted or region-restricted
- the URL points to a playlist or wrapper page rather than one accessible video
- the external service cannot retrieve that source
If permitted, obtain the best local copy and use file-based keyframe extraction instead. Keep the original post URL and context in your notes.
Why InVID Reverse Search Sometimes Finds Nothing
No-match results are common and should be interpreted conservatively.
The video is genuinely new or poorly indexed
The video is private, deleted or from a closed platform
The selected keyframe is too generic
The clip was heavily transformed
The important scene was not in the initial keyframes
Do Not Upload Sensitive Evidence Without Checking Policy
For confidential, legal, corporate or highly sensitive media, check your organization’s data-handling rules before submitting a file or URL to any external verification service.
The InVID project notes that some plugin functions use external services. That is normal for a browser-based verification toolkit, but it matters when the source media contains confidential information, personal data or restricted evidence.
For sensitive cases, preserve the original independently and document which external services were used.
A 10-Minute InVID Verification Routine
- Capture the claim. Save the URL, caption, uploader and stated event.
- Run the Keyframes tool. Use the URL or local file.
- Pick three visually different frames. Prioritize context, not just faces.
- Reverse search each frame. Open the actual result pages.
- Request more frames if needed.
- Compare dates. Look for an occurrence that predates the current claim.
- Find the longest credible version. Check source credits and context.
- Search visible text and watermarks separately.
- Write a narrow conclusion. Earlier copy found, source supported, context contradicted, or unresolved.
A Better Verification Note Template
Use Precise InVID Verification Verdicts
| Verdict | Meaning |
|---|---|
| Earlier occurrence found | The same visual was publicly available before the current post or claimed event |
| Longer source found | A longer version provides additional context around the viral excerpt |
| Source strongly supported | Visual match, attribution, chronology and publisher evidence point to the same source chain |
| Current context contradicted | Earlier or authoritative evidence conflicts with the date, location, event or description attached to the viral post |
| No useful visual match | The searched frames did not produce a useful indexed result |
| Source unresolved | Available keyframes and search results are insufficient to establish a reliable source history |
Where InVID Fits in a Complete Video Verification Process
InVID is strongest at the source-discovery stage.
A complete investigation may still need to verify:
- who uploaded the earliest credible copy
- where the footage was recorded
- when it was recorded
- whether the video was edited
- whether AI-generated media is involved
- whether the caption or real-world claim is accurate
For that broader decision process, use the video verification guide.
Key Takeaway
InVID verification is most useful when you treat keyframes as investigative queries, not automatic answers.
Extract several visually distinct frames, reverse search the strongest ones, request more keyframes when the first set is weak, open the result pages, compare dates and follow source credits backward. An older match can disprove a “breaking news” claim, and a longer source can restore missing context.
The tool accelerates verification. The final judgment still comes from source comparison, chronology, context and corroborating evidence.
FAQ About InVID Verification
What is InVID verification?
InVID verification is a browser-based fact-checking workflow that helps users extract video keyframes, run reverse image searches, inspect context and metadata, and trace images or videos to earlier sources.
Is InVID the same as InVID-WeVerify?
The verification plugin began with the InVID project and was later expanded through WeVerify and vera.ai. The current Chrome extension is listed under the InVID WeVerify VeraAI name, while many users still refer to it simply as InVID.
Is InVID free?
The public browser extension is available free of charge. Some advanced functions or external services may have access conditions, registration requirements or operational limits.
How do I reverse search a video with InVID?
Open the Keyframes tool, submit a supported video URL or local file, wait for representative frames, then reverse search several strong keyframes and investigate the pages returned by the search engines.
What is the best InVID keyframe to search?
Choose a clear frame with distinctive contextual information such as signage, landmarks, logos, watermarks, buildings, vehicles or unusual objects. Search several visually different frames rather than relying on one.
What should I do if the first InVID keyframes return no matches?
Try other frames, request more keyframes where available, search visible text or watermarks, or use a different reverse image search engine. No match does not mean the video is authentic or original.
Can InVID find the original video automatically?
No. It can help locate matching and earlier media, but the investigator still needs to compare publication dates, attribution, video length, source credibility and earlier references to determine the strongest source history.
Can InVID detect deepfakes?
InVID includes verification and still-image forensic tools, but its keyframe workflow is primarily a source-tracing method. Deepfake detection is a separate technical task and should not be inferred from whether a reverse image search succeeds or fails.
Can I upload a video file to InVID?
The InVID keyframe service supports local video-file input in addition to supported URLs. For sensitive material, review applicable privacy and evidence-handling requirements before using external services.
Why does InVID find an old copy but not the original?
Search engines only expose media they can index. The oldest result you find may itself be a repost. Treat it as an earlier known occurrence and continue following attribution and source clues backward.