In this post, you’ll get a quick tour of the latest Amped Authenticate release, including updates to diffusion-model deepfake detection, improvements to geometrical analysis, the new Inspector panel, and workflow enhancements across Image and Video Mode.
Dear friends, we’re excited to share that a new update of Amped Authenticate has just been released! Following the feedback from our wonderful users and the recent developments of AI generation models, we’ve brought several improvements to the software:
- We’ve updated the Diffusion Model Deepfake filter with many new models and significantly sped up the embedded batch processing feature;
- We’ve improved all filters in the Geometrical Analysis category by allowing to quickly shrink or enlarge constraints while monitoring the system’s feasibility;
- We added the new Inspector panel;
- We improved the user interface;
- We added new navigation capabilities to the Video Mode;
- …and more!
There is so much to see, but it’s always worthwhile reminding ourselves why image and video authentication and deepfake forensics require a multi-faceted approach and toolset. As you will see as we walk through the news of this release, with Authenticate you can cross-check your findings with diverse and complementary tools. This helps you reach strong and defensible conclusions.
Let’s dive into the details!
Key Takeaways
- You can process large evidence folders faster thanks to the updated Diffusion Model Deepfake filter, with added models and accelerated embedded batch processing.
- You get better traceability for deepfake classifications with a new Detection Model Version column in the results table.
- You can build more defensible, explainable findings with strengthened Geometrical Analysis, including the ability to quickly shrink/extend constraints while watching the feasibility results update in real-time.
- You can verify details faster with the new Inspector panel in both Image and Video Mode, showing pixel values under your cursor and supporting quick copy to clipboard (CTRL+C).
- You’ll work more efficiently with UI upgrades like closable panels in Image Mode, a one-click maximize Viewer option, and improved Filters panel hover/description behavior.
- You can move through footage more systematically with new Video Mode frame navigation (custom frames, I-frames, or time units) and improved Image/Video Mode integration when sending frames for analysis.
- You benefit from additional performance, compatibility, and stability improvements (e.g., AV1 support via libdav1d, updated Exiftool/IPP, plus multiple fixes).
See the New Features in Action
Updated Diffusion Model Deepfake Filter
The Purpose of the Diffusion Model Deepfake Filter
If you’re monitoring the evolution of deepfake generators, you know they’re making astonishing progress very quickly. For this reason, it’s paramount that newly released AI models are added to the training set of our AI-powered Diffusion Model Deepfake filter.
We’ve said that since the first release of the filter, and we’ll reiterate it here once more: this is a machine learning-powered detector. So, it’s not meant to be used as the sole indicator to label an image as being deepfake, let alone real. Deepfake forensics is much more than deepfake detection; that’s our motto. Nevertheless, a powerful detector can be so helpful when it comes to an initial screening, especially when you’re confronted with hundreds or thousands of images.
Updates to the Machine Learning Model
We’ve included images from many new diffusion models, both old and new, including the latest release of Nano Banana 2, which was released just a few days ago!
When training our detector, we conduct data augmentation to increase the detector’s robustness to post-compression, downscaling, and other transformations. Despite all our efforts, you may still find situations where the filter returns a false positive (a real image being flagged as compatible with an AI model) or a false negative (an AI-generated image being flagged as not compatible with a known AI model). This is not to be considered a “bug”, we simply have to accept that no perfect detector exists.
To make it clear to the user that the results from this filter are obtained with an AI system, we’ve also added a reminder within the GUI. This applies to the Face GAN Deepfake and the Fusion Map filters as well.
Analyze All Images in Evidence Folder: Now Much Faster
The title says it all! If you have to run Diffusion Model Deepfake on a folder of images, we recommend doing it using the dedicated button in the filter’s interface. That will make it run much faster than processing each image individually.
We’ve also added a column to the table showing the Detection Model Version used to carry out the classification, for improved traceability of results.
Improvements to Geometric Analysis Filters
The Advantages of Geometrical Analysis
Geometric analysis sits on the opposite side of machine-learning-powered detectors. It takes longer to use it since it requires user input, but it rewards you with total robustness and explainability.
- Total robustness: you can meaningfully apply a shadow, reflection, and perspective analysis even on an image that you know has been printed and recaptured, or on a video frame. Something that would fool even container-based analysis, but that it’s fully captured by geometrical analysis.
- Explainability: when something is off, it’s very easy to explain why, possibly even to the layman. Concepts like shadows and vanishing points do not call for an engineering degree to be grasped!
New Ability to Shrink or Extend Constraints
When using filters in the Geometrical Analysis category, a key element helping the explainability is always setting constraints that are “indisputable”. Take this image (generated with the brand new Nano Banana 2 and cropped to remove the Gemini logo).
Shadows look great to the naked eye. However, using the Shadows filter, we rapidly identify the following constraints, which reveal the system is unfeasible (which means: shadows are inconsistent).
Do we really need all those constraints to show that the shadows don’t work out well? Probably not! Luckily, Authenticate will figure that out for us: let’s click on the Show Minimal Unfeasible Set button and we’ll see that three constraints alone are enough to make the system unfeasible.
Having fewer constraints reduces the room for discussion with whatever opposing party. But with this Authenticate update, we’re helping you even further.
In the example above, the constraint associated with the airplane’s front wheel is a bit tight at the moment, as shown in the red rectangle below. Someone could argue that, if you select the entire wheel, then maybe the shadows would turn out to be consistent.
Would the system remain unfeasible if we had done a more conservative selection?
Well, we can now check it very quickly! We just hover with the mouse on the wheel’s constraint in the right panel. Then, we hold the CTRL key and move the mouse wheel to extend or shrink the constraint. While we do that, the system’s feasibility label will update simultaneously.
In this way, we can enlarge the constraint so that the wheel is entirely contained. We can then check whether the system remains unfeasible, which happens to be the case here: the expanded constraint shown below leaves way less room for discussion!
New Inspector Panel and Improved GUI
The New Inspector Panel
We’ve added a new Inspector panel to both Image and Video Modes! The presented information shows you the value of pixels in the viewer while you move the mouse over them. With a simple CTRL + C, the information will be copied to the clipboard so that you can paste it elsewhere, e.g. in a bookmark’s description.
This could be useful, for example, to highlight the presence of saturated pixels, that pose a challenge to some forgery localization algorithms. It can also help compare the values of individual pixels across a Reference and Evidence image.
Ability to Close Panels in the Image Mode
We also added the ability to close panels in the Image Mode, as you could already do in the Video Mode, and re-open them from the View menu.
Ability to Maximize the Viewer Panel
Whenever you need more room to focus on a filter’s output, there’s a new button to temporarily maximize the Viewer panel and hide all the rest. Click again on the same button to go back to the previous view.
Filters Panel: Added Hovering Effect and Description
We’ve improved the Filters panel by adding a hovering effect when you move the mouse over filters, so you know what you’re about to select!
In addition, the Image Mode will now show you a brief description of each filter as you hover over it, something to which Amped FIVE users are familiar.
If you consider yourself a seasoned user who doesn’t need this aid, you can turn it off from the Program Options, General tab.
Improved Integration Between the Video and Image Mode
Authenticate Video Mode offers you the ability to send an individual frame to the Image Mode for further analysis. Of course, you should keep in mind that not all Image Mode’s filters are meaningfully applied to a video frame. For example, checking the File Format wouldn’t make much sense since the file is a new PNG obtained by saving the video frame. On the other hand, filters in the Geometrical Analysis can certainly be used on individual video frames.
To help the user identify which filters should not be used, the Image Mode will now display a dialog and then disable them automatically when receiving a video frame from the Video Mode. The user can still re-enable any filter by right-clicking on its name in the Filters panel, as usual.
The Video Mode now offers more ways to search through the video thanks to a dedicated addition to the Player panel. You can navigate by a custom number of frames, I-frames, or temporal units. This can be very handy when you want to systematically “probe” video frames to see if or when something happens.
Other Improvements and Bug Fixes
We’ve made other improvements to the software, including:
- Improved performance of the Clones Blocks filter
- Improved performance of the JPEG QT filter when loading the QT database
- Added the libdav1d library to FFMS to allow loading AV1 videos
- Updated Exiftool to v. 13.44 and IPP to 2025.3.0.371
- Avoid showing the “Frame count mismatch” warning dialog for interlaced videos
- Fixed some bookmark numbering issues in the report
- Fixed a bug that caused the Clear Evidence File button not to work in some cases
Don’t Delay – Update Today
The new features make Authenticate more powerful than ever.
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.
FAQ – Authenticate Update 40165
In this release, you get an updated Diffusion Model Deepfake filter, stronger Geometrical Analysis tools, a new Inspector panel, UI improvements in Image Mode, new navigation options in Video Mode, and additional performance, compatibility, and bug-fix updates.
It helps you analyze image sets faster and with broader coverage. The filter now includes many new diffusion models, speeds up embedded batch processing, and adds a Detection Model Version column for better result traceability.
No. You should not use a machine-learning detector as the only basis for calling an image fake or real. In the release notes, Amped explicitly explains that no perfect detector exists and that deepfake forensics requires a multi-faceted approach.
You can now quickly shrink or extend constraints while monitoring feasibility in real time. That makes it easier for you to test more conservative selections and build findings that are both sound and easier to explain.
It gives you a more explainable and robust way to assess evidence. Geometrical analysis can still be meaningfully applied even to printed-and-recaptured images or video frames. It’s often easier to communicate the result using visual cues like shadows, reflections, and perspective.
The new Inspector panel shows you pixel values as you move your cursor over the viewer in both Image and Video Mode. You can also copy that information with CTRL+C, which is useful when documenting observations such as saturated pixels or comparing values between reference and evidence images.
You can now close panels in Image Mode and reopen them from the View menu. There is also a new button to temporarily maximize the Viewer panel, plus improved hover behavior and filter descriptions in the Filters panel.
You can now navigate video by a custom number of frames, I-frames, or temporal units, which helps you probe footage more systematically. We also improved the handoff from Video to Image Mode by warning you about filters that do not meaningfully apply to extracted video frames and disabling them automatically.