How Investigators Should Handle
AI-Enhanced Video Evidence

The Investigator's AI Video Challenge

Investigators of every type — private investigators on personal injury matters, corporate security teams handling internal theft investigations, insurance fraud investigators reviewing claims, background check specialists, and law enforcement agencies — work with video that rarely arrives in ideal condition. Surveillance cameras age and are never upgraded. Dashcams record at low resolution with aggressive compression. Ring doorbells and home cameras produce footage that is technically adequate for casual review but barely usable for close identification. ATM cameras and cash register cameras are optimized for minimal storage cost, not image quality.

AI video enhancement has created real value in this context. Footage that previously required expensive forensic laboratory analysis — or was simply presented as "best available evidence" with apologies for its quality — can now be processed in minutes on a professional workstation. Faces become identifiable. License plates become readable. Important details that were lost to compression become visible. The investigator's ability to produce useful evidence from difficult footage has improved dramatically.

But with that capability comes a responsibility that many investigators are not yet prepared for: the responsibility to handle AI-enhanced evidence with the same rigor applied to any other forensic tool. An investigator who runs surveillance footage through an AI tool, downloads the enhanced version, and presents it to a client or attorney as "the footage" — without preserving the original, documenting the process, or understanding the evidentiary implications — has created a problem that may not become apparent until the matter reaches litigation.

Understanding What AI Enhancement Does and Does Not Do

Before discussing how to handle AI-enhanced evidence, investigators must understand what the technology actually does — because the explanation may need to be given to attorneys, clients, and eventually courts.

AI video enhancement uses trained neural network models to predict and reconstruct detail in degraded video frames. Real-ESRGAN, the model type used in BetterVideo, is a super-resolution model: it takes a low-resolution frame and generates a higher-resolution version by predicting what additional detail should be present based on patterns learned from training on thousands of high/low resolution video pairs. GFPGAN is a face restoration model: it identifies facial regions and applies a restoration process that improves facial detail.

The critical distinction to understand is between enhancement and fabrication. Proper AI enhancement tools improve the clarity of detail that was present in the original scene but degraded during capture. They do not insert people, objects, or events that were not present. However, they do make probabilistic predictions about what degraded pixels should look like — meaning the enhanced output represents the AI's best reconstruction of the original scene, not a direct amplification of recorded data. This distinction matters for how you explain the footage and what claims you can make about it.

Enhancement can make a blurry face more recognizable, but it cannot identify a face that was genuinely not captured. Enhancement can sharpen a license plate that was captured but poorly compressed, but it cannot reconstruct a plate that was never in frame. When presenting enhanced footage, investigators must be accurate about what the original footage contained versus what the enhancement has clarified.

The Investigation Evidence Protocol for AI Enhancement

The following protocol should be applied consistently by every investigator who uses AI video enhancement on footage that may be used in legal or administrative proceedings.

Step 1 — Secure and document the original: Before any enhancement, create a forensic copy of the original footage. Compute the SHA-256 hash of the original file and record it in the case file with the date, time, case number, and who performed the hash computation. Store the original on write-protected media or in a tamper-evident storage system. From this point forward, the original is untouchable — it will be produced as-is at any stage of proceedings.

Step 2 — Create the enhancement record: Document the enhancement session before you begin: date and time, case number, file being processed (referenced by hash), enhancement tool name and version, and who is performing the enhancement. Record the specific settings or options selected. For BetterVideo, record whether 1080p or 4K enhancement was selected and the AI models applied (Real-ESRGAN x2plus for upscaling, GFPGAN v1.4 for face restoration).

Step 3 — Perform the enhancement: Upload the original footage to the enhancement tool. After downloading the output, immediately compute the SHA-256 hash of the enhanced file and record it. Record the download date and time.

Step 4 — Store and label: Store both original and enhanced files with clear, unambiguous labels. The original file should be labeled with "ORIGINAL — DO NOT MODIFY" and the enhanced file labeled "AI ENHANCED — [date] — [tool name and version]." Store in a case management system that logs access.

Step 5 — Prepare the explanation: Write a brief technical explanation of what the enhancement tool is, what it does, and what changes it made to the footage. This should be understandable to a non-technical attorney or client. Include references to the AI models used and their published technical documentation if available.

Step 6 — Disclose to the engaging party: When delivering findings to the client or retaining attorney, disclose that AI enhancement was used, provide both the original and enhanced versions, and include the enhancement documentation. The retaining attorney needs this information to decide how to use the footage and whether and when to disclose to opposing parties.

Privacy Obligations When Handling Surveillance Footage

Beyond the evidentiary considerations, investigators handling video footage have privacy obligations that AI enhancement does not change — but that AI enhancement can make more acute. AI enhancement can reveal identifying detail that was not visible in the original footage — making unidentifiable individuals identifiable, or making redacted license plates legible. This increased resolution of identifying detail creates additional privacy considerations for how the footage is handled and shared.

In many jurisdictions, surveillance of individuals in private locations is restricted or prohibited. Recording in certain locations (healthcare facilities, legal offices, financial institution interiors) may be subject to additional restrictions. Biometric data laws in states like Illinois, Texas, and Washington may impose requirements on how facial recognition data — including AI-enhanced facial detail — is collected, stored, and used. Investigators should review the applicable privacy law in their jurisdiction before enhancing footage that may contain biometric data.

For footage handled on behalf of insurance companies, financial institutions, or healthcare organizations, additional privacy frameworks (HIPAA, GLBA, state insurance regulations) may apply. The engagement agreement with these clients should address how footage is handled, including restrictions on third-party tools used for analysis.

Common Mistakes Investigators Make with AI-Enhanced Evidence

• Overwriting the original: Processing footage and saving only the enhanced version. Without the original, there is no way to demonstrate what changed and no foundation for the enhanced version's credibility.
• Using consumer tools without documentation: Running footage through a free online tool, downloading the result, and presenting it without any record of what tool was used, when, or what it does. This version of enhanced footage is extremely vulnerable to evidentiary challenge.
• Uploading to platforms with inappropriate terms: Using platforms that train AI models on uploaded content, retain footage indefinitely, or claim broad license rights — potentially exposing confidential investigation footage to unauthorized use.
• Not disclosing enhancement to counsel: Presenting enhanced footage to an attorney without disclosing that it was AI-enhanced. The attorney needs to know before deciding how to use the footage in proceedings.
• Claiming the enhanced version shows things the original doesn't contain: Enhancement clarifies degraded detail; it does not create new information. Investigators should never claim that an enhanced version reveals something that was not captured at all.
• Treating enhanced and original as interchangeable: Both versions have evidentiary value and serve different purposes. The original is what was recorded; the enhanced version is what analysis tools reveal. Present both, clearly labeled.

BetterVideo Features That Support Investigation Evidence Handling

BetterVideo stores original and enhanced video separately, with both files accessible from the dashboard for the 30-day retention window. The original is never modified — enhancement produces a distinct output file. This architecture supports the chain of custody requirement that original and enhanced versions be separately maintained.

BetterVideo does not train AI models on uploaded footage, which eliminates the risk that investigation footage is embedded in a commercial AI system. Processing uses named, documented AI models (Real-ESRGAN x2plus, GFPGAN v1.4) with published technical specifications that investigators can reference in their enhancement documentation. The 30-day retention window provides adequate time for legal and administrative processes while ensuring footage is not retained indefinitely on vendor infrastructure.