This fall’s Authenticate update brings great improvements to our deepfake detection features, with the ability to run deepfake detection on batches of images and from within the Smart Report!
Dear friends, here we are again, presenting a new update to Amped Authenticate! As usual, we’ve been inspired by the requests coming from you, our loved users. There’s a lot to show:
- we’ve improved the user interface,
- we’ve simplified the Diffusion Model Deepfake filter and added a batch processing feature for it,
- we’ve added a new Deepfake Detection category into the Smart Report tool,
- and more!
A lot of news, so let’s dive in and check them all out!
See the New Features in Action!
Graphical User Interface Improvements
Dockable Panels for Image Mode
Our engineers improved the GUI consistency between Authenticate’s Image and Video Modes. You can now undock panels in the image mode, so as to organize the GUI as it suits your working habits better. If you get lost, you can always restore the default windows layout from the View menu.
For users who upgraded from a previous version: if undocking doesn’t work for you, simply click on View – Reset Windows Layout to fix the issue.
New Option for Loading Files in Visual Inspection
Until today, when you dragged a new file into Authenticate or loaded it from the Evidence panel, the currently selected filter would automatically be run on the file. This can be useful in some cases, but could waste time in others.
Starting with this release, you’ll be able to choose the default behavior! Head to the Program Options and decide whether newly loaded files should be automatically opened in the Visual Inspection filter, or in the currently selected filter. This applies to both the Image and Video Modes.
Please note that this option does not affect the behavior of the “next/previous file” buttons, as they are conceived to allow a quick comparison of the current filter’s results among images in the same folder.
Seeking a Specific Frame for Video Mode
The Video Mode lets you now seek a specific frame by typing its number in the Viewer and hitting Enter. This is very handy when you’re hunting for a specific frame in a long video.
Improvements to the Diffusion Model Deepfake Filter
Diffusion Models are the deep learning technology behind the most recent deepfake generators. They create images, and nowadays videos, that are very hard to distinguish from real content with the naked eye.
Authenticate’s Diffusion Model Deepfake filter is a Machine Learning (ML) based tool trained to distinguish natural images from synthetically generated images created with tools like Dall-E, Midjourney, Stable Diffusion, and Flux. Like every ML tool, it can be wrong sometimes, which is why we always recommend cross-checking its results with other types of analysis, e.g. looking for geometrical inconsistencies, visual artifacts, and so on.
A Brand New Dress
With this release, we’ve improved and simplified the way the Diffusion Model Deepfake filter provides its results. Instead of an output table showing compatibility scores for each of the inner classes, we’ve narrowed down the output to two scores:
- compatibility with a known diffusion model
- non-compatibility with a known diffusion model
This makes interpreting the output much easier.
Moreover, we’ve turned the output into a graphical representation. If the image is found to be compatible with a diffusion model, you’ll see a red border around it, along with the compatibility score displayed at the bottom.
Instead, when the image is classified as not compatible with a known AI model, Authenticate will draw a more neutral gray border around it. This indicates that the image’s trustworthiness cannot be determined solely from that result. The image could be fake, even if AI didn’t generate it, or it may have been created by an AI system that the filter could not detect.
You can still access results in a tabular form by clicking on the dedicated button in the filter GUI for easier copy-pasting.
Batch Processing
We’ve also introduced the ability to run the Diffusion Model Deepfake filter on a folder of images.
Results are presented in a very convenient table from which you can right-click to load images for further analysis. As for any other table in Authenticate, you can right-click on it to export results to an HTML or TSV file.
Face GAN and Diffusion Model Deepfake Filters Now Available in Smart Report
A Quick Introduction to the Smart Report
Authenticate’s Smart Report is an excellent way to do an initial triage of your images. For those unfamiliar with it, the Smart Report analyzes the image’s file format and metadata, looking for suspicious signs of processing (e.g., absence of metadata, non-standard Huffman Tables, and more).
- If all looks good, the image is marked with a green light as being “likely camera original”, keeping everything very quick.
- If, instead, at least one warning is raised, then a set of local analysis filters is run on the image. If a local analysis filter raises a warning, the image gets flagged with a red light as containing “traces of possible forgery”.
- Otherwise, it goes into the intermediate state, marked with a yellow light.
You can read more details about the Smart Report here.
Running Deepfake Detection in the Smart Report
With this release, we’re introducing the ability to run the Face GAN Deepfake and/or the Diffusion Model Deepfake filters as part of the Smart Report. This is an optional feature that you can toggle from the Smart Report configuration dialog.
You’ll find the results from the Deepfake Analysis filters in a dedicated section, below the Local Analysis filters results.
We remark that the Smart Report is to be considered a triage tool. It is obviously not meant to replace a thorough analysis of each image.
Input File Details in the Report
Providing details and hash values for input files is a must-have for every forensic report. Since the very first release of Authenticate projects, every bookmark in the report has always featured the hash value of the analyzed file.
However, for large projects involving many files, users may find it more convenient to group all hash values and more details about each input file at the top of the report. That’s exactly what we’ve introduced in this release! In the Generate Report dialog, both for Image and Video Modes, you can now enable the new option “Add input files hash and info”.
Of course, each input file will only be listed once, even if you have multiple bookmarks for it. This is what you’ll find in your report:
Improvements to the Macroblocks and Coding Tree Units Filters
We have considerably improved how you can display data in the Macroblocks and Coding Tree Units filters!
First of all, you can now customize the color of the subdivision lines and of each motion vector type, which helps a lot when working with very bright videos.
We’ve also improved the way the Quantization Parameter value is displayed. First of all, for larger coding tree units, the value will be displayed only once at the center instead of being repeated multiple times.
Moreover, when visualizing the Quantization Parameter as a color scale, a new Color Map parameter is available. By setting it to “Frame-based”, the color scale will be determined on a relative basis, according to the maximum and minimum quantization parameter values within each individual frame. By setting it to “Absolute”, the minimum and maximum values allowed by the codec will be used instead.
In the example below, you can see that even a subtle difference like the one between 25 and 19 becomes noticeable when the Color Map is set to Frame-based.
Other Improvements and Bug Fixes
We’ve made other improvements to the software, including:
- Annotate: added the ability to copy-paste the position of annotation objects.
- Annotate: Edge Feathering now adjusts proportionally to the redaction area.
- Improved the description of the Face GAN Deepfake, Diffusion Model Deepfake, and Fusion Map filters.
- GUI: Fixed the appearance of losing colors during switching users due to Windows behavior.
- Advanced File Info: fixed a bug causing incorrect value being of “PTS duration (computed)” for some peculiar video formats.
- Uninstaller: the license seat can now be deactivated while uninstalling manually if the Digital license scheme is in use.
- Video Mode: Advance File Info: added the ability to display the desired frame in the viewer via “Go to Frame”.
- Video Mode: fixed a bug that caused the chroma upsampling option not to be applied correctly.
- Added a warning message that notifies the user when some software components cannot be found or accessed.
Don’t Delay – Update Today
The new features we’ve added can significantly speed up your workflow and help you focus on the most suspicious images first.
If you have an active support plan, you can update straight away by going into the menu About > Check for Updates within Amped Authenticate. If you need to renew your SMS plan, please contact us or one of our authorized distributors. And remember that you can always manage your license and requests from the Amped Support Portal.