Looking to implement C2PA? Trufo provides tooling to take care of everything from certificates and timestamping to watermarking and fingerprinting. Learn More
Trufo wordmark
A New Gold Standard for Watermarks

Introducing PawPrint

The first neural watermark designed for provenance at scale.
Published August 30, 2026; Updated September 4, 2026

We are introducing our new line of AI-powered watermarks, PawPrint, available for image and audio and video. PawPrint represents a massive leap in performance over existing models, across all dimensions (speed, accuracy, durability, perceptibility, etc.). It can be swiftly deployed in any business use case where digital media content needs to be reliably identified. PawPrint is available today as part of our C2PA Signing API & SDK services.

Overview

Trufo's pawprint.video watermark delivers a substantial improvement in watermark performance across the board:

WatermarkCompanyYearPayload [bits]Decode [ms]VMAF ↑FPR ↓Durability ↑Score
pawprint.videoTrufo202644A-1.55A95.06A-0.000A0.780A-94
pixelseal.bch44²Meta202544A-2.62A-96.71A-0.000A0.583B-86
videoseal.bch128²Meta2024128A2.57A-95.44A-0.000A0.530C+86
rivaganAcademic201932B+5.41B+98.68B+1.000D0.313C-56
1Assigned grades are adjusted based on a mix of other perceptibility metrics.
2Enhanced by Trufo with BCH error correction.

When paired with Trufo's provenance platform, pawprint.video becomes the natural choice for practical watermark deployment in almost any commercial setting.

Fluid and per-Frame

The pawprint.video watermark is applied to every frame, creating a fluid watermark that is imperceptible (adversarially-trained) and robust (even on screenshots). Here is an example of the watermark:

Original
Watermarked

To visualize what the watermark looks like, here is the watermark residual, at 24× amplification, of the four watermarks we tested.

pawprint.video · ×24
pixelseal.bch44 · ×24
videoseal.bch128 · ×24
rivagan · ×24

The pawprint.video watermark offers a good balance of luma and chroma components, and moves dynamically with the underlying video content.

Evaluation

We compare pawprint.video against three video watermarks that we were able to obtain for testing:

  • pixelseal (Meta, 2025)
  • videoseal (Meta, 2024)
  • rivagan (Academic, 2019)

The evaluations are conducted on a collection of 100 10s clips drawn from four datasets: wikimedia-commons, nasa-svs, blender-open-movies, and xiph-pd. The test set was excluded from the training set.

Speed

We benchmarked encode and decode speed, from compressed input file to compressed output file, for full-length clips.

pawprint.videopixelseal.bch44videoseal.bch128rivagan
encode (fps)178.6226.1231.3148.8
full decode (fps)644.5381.0388.4185.0

Overall, pawprint.video is the fastest, both in decoding and (once correcting for how pixelseal and videoseal copy watermarks across frames) in encoding.

Perceptibility

We looked at five standard metrics: PSNR, SSIM, VMAF, LPIPS, and wM4E.

WatermarkPSNR [dB] ↑SSIM ↑VMAF ↑LPIPS ↓wM4E ↓
pawprint.video44.980.9809295.060.004500.271
pixelseal.bch4447.120.9565396.710.003560.759
videoseal.bch12846.970.9879995.440.003120.413
rivagan42.520.9402998.680.044541.243

They present completely conflicting opinions; the takeaway is that quantifying perceptibility is a tricky matter and overall all four watermarks are solid.

Durability

We tested exact recovery of the payload against a variety of attacks. The evaluation included the following attacks:

  • compression (H.264 CRF 20-30)
  • reflection & rotation (orthogonal)
  • resolution changes (720p, 360p) & aspect ratio changes
  • speed modulation
  • grayscale & color filters

Note that we set the "clean" baseline to be 1080p CRF 18. Three types of evaluations were done: on single-frames, on 1s-segments, and on full-length clips.

WatermarkPayload [bits]Clean frameClean 1 sClean 10 sAttacked frameAttacked 1 sAttacked 10 sFPR
pawprint.video4491%99%100%73%77%84%0%
pixelseal.bch444453%90%98%39%64%72%0%
videoseal.bch12812858%94%99%32%58%69%0%
rivagan3278%85%91%27%33%34%100%
clean, by read length
0.000.250.500.751.00pawprint.videopixelseal.bch44videoseal.bch128rivagan
frame
1s
10s
attacked, by read length
0.000.250.500.751.00pawprint.videopixelseal.bch44videoseal.bch128rivagan
frame
1s
10s

Trufo's pawprint.video shows by far the highest durability, across all attack types (not shown here), and especially in the single-frame setting (e.g. screenshots).

Overall Score

The overall ranking of tested watermarks, based on our composite score, is as follows:

pawprint.video
94
pixelseal.bch44
86
videoseal.bch128
86
competitor est.
80
legacy est.
60
rivagan
56

Most video watermarking systems are not public, so we cannot easily test them. Based on our knowledge, we estimate that traditional commercial systems score around 60 while those with strong AI/ML teams score around 80.