VideoCCTV / DVR AnalysisExport footage from Hikvision, Dahua and other DVRs with correct times, frames and hashesDeepfake / AI Forensics
Examine face swaps and AI-made media, check C2PA provenance and know where detectors fail
- Advanced
- Online or on campus
- 5 hands-on labs
- 7 modules
About this course
A video of a public figure saying something he never said. A voice note from a 'boss' asking for an urgent bank transfer. A morphed photo used to blackmail a college student. Synthetic and manipulated media now reach cyber cells and courts. Someone has to give an honest opinion on them.
This course teaches you to form that opinion. You learn how face swaps, lip-sync models, image generators and voice clones work, because knowing how something is made tells you where it fails. You examine visual artefacts such as blending edges, mismatched lighting and reflections, teeth and ear detail and flicker between frames. You also learn classic image forensics such as error level analysis, noise analysis and double compression checks, along with their real limits.
Then you look at provenance. C2PA Content Credentials, invisible watermarks and tracing a file back to its first upload can sometimes show where it came from, which is often stronger than any artefact. The course is direct about detection tools. A detector score is a probability from a model trained on certain data. It can be wrong in both directions and it is not proof on its own, so your conclusion has to rest on several independent checks.
Face-swap artefactsError level analysisC2PA provenanceDetector limits
After this course you can
- Explain how face-swap, lip-sync, image generation and voice cloning methods work and where each tends to fail
- Examine a suspected deepfake frame by frame for blending, lighting and temporal artefacts
- Run error level analysis and noise checks and explain why their results can mislead
- Read and verify C2PA Content Credentials and explain what a missing manifest does and does not mean
- Use detection tools as one input among several and describe their limits in your report
- Write a balanced opinion that combines source, metadata, provenance and visual analysis
Syllabus
7 modules · 28 topics
01How synthetic media is made4 topics
- Face swaps built with autoencoder tools such as DeepFaceLab and FaceSwap
- Lip-sync and face reenactment models
- GANs and diffusion models for fully generated images
- Voice cloning and text-to-speech
02Visual artefacts5 topics
- Blending boundaries and skin texture or colour mismatch around the face
- Lighting, shadows and eye reflections that do not agree
- Teeth, ears, jewellery, hair edges and hands
- Flicker and warping between frames, especially when the head turns
- Lip movement that does not match the speech sounds
03Classic image forensics and its limits4 topics
- How error level analysis works and why it lights up edges and textures in genuine photos
- Why ELA tells you little about files recompressed by social media or saved as PNG
- Noise analysis and sensor pattern noise (PRNU)
- Double JPEG compression and copy-move detection
04Audio deepfakes3 topics
- Reading spectrograms and what to look for
- Breathing, room sound and unnatural pauses
- Why short, compressed voice notes are hard to judge
05Provenance and source5 topics
- C2PA Content Credentials, signed manifests and the edit history they record
- Checking a manifest with c2patool and the Content Credentials verify site
- Why a missing manifest proves nothing and a valid one shows only what its signer recorded
- Invisible watermarks such as Google SynthID
- Reverse image search and finding the earliest copy of a file
06Detection tools4 topics
- How detectors are trained, with datasets such as FaceForensics++ and DFDC
- Why detectors struggle with new generators and compressed video
- False alarms on real footage and the 'liar's dividend', where genuine media is called fake
- Using detector scores responsibly
07Opinion and reporting3 topics
- Combining source, metadata, provenance and visual evidence
- Scales of opinion and wording that does not overstate
- Presenting findings to investigators, lawyers and courts
Hands-on labs
- Lab 1
Examine a face-swap video frame by frame and log every artefact with its frame number
- Lab 2
Run ELA on original, edited and WhatsApp-forwarded images and compare what it shows
- Lab 3
Verify C2PA Content Credentials on a signed image, then strip them and check again
- Lab 4
Run two detection tools on the same clips and explain where they disagree
- Lab 5
Write a short opinion on a suspected morphed photo using at least three independent checks
Tools you will use
- Amped Authenticate
- c2patool
- InVID-WeVerify plugin
- ExifTool
- FFmpeg
- Audacity
- Python with OpenCV
- Reverse image search engines
Who it suits
- Image and video examiners in forensic laboratories
- Cyber cell officers handling morphing and impersonation complaints
- Journalists, fact-checkers and trust and safety teams
- Lawyers who need to question or defend media evidence
Before you join
- The Image & Video Metadata course or similar experience
- A basic idea of how digital images and video compression work
- Some Python helps but is not required
Questions about this course
Is there a tool that tells you for sure if a video is a deepfake?
No. Detectors give a score from a model and that score can be wrong, especially on new generators or heavily compressed video. A sound opinion combines several checks and the course teaches you how to do that.
Does ELA prove a photo has been edited?
No. ELA shows how parts of a JPEG respond to being saved again. Edges, textures and earlier recompression all change the result, so bright areas are not proof of editing. Treat it only as a pointer to look closer.
What if the image carries C2PA Content Credentials?
A valid manifest shows which device or software signed the file and what edits it recorded. That is useful. It is still the signer's record and not a check on what the picture shows. A file without credentials is not suspect for that reason alone.
More courses in Surveillance & Image / Video Forensics
VideoCCTV / DVR AnalysisExport footage from Hikvision, Dahua and other DVRs with correct times, frames and hashes
VideoImage & Video MetadataRead EXIF, XMP, GPS tags and MP4 atoms with ExifTool and know what each field can proveNot sure this is the right one for you?
Tell us your background and what you want to do next. A counsellor will suggest a course and the next batch.