How to Make A Deepfake Video

The term “deepfake” has become a famous (and infamous) concept.

Combining “deep learning” with “fake,” deepfakes are AI images and videos created via a deep learning method known as GANs - generative adversarial networks.

Have you noticed how easy it is to create deepfakes these days? With just a few clicks, anyone can whip up convincing fake images, videos, or even audio recordings.

While the technology behind deepfakes, like Generative Adversarial Networks (GANs), is pretty complex, most of the consumer tools out there simplify the process so you can focus on creating rather than coding.

This ease of use raises some serious concerns about how this technology could be misused, from spreading misinformation to creating harmful content. As deepfake technology continues to evolve, it’s essential to understand both its potential and the challenges it brings to our digital landscape.

But it’s not all negative.

For product marketers, using a deepfake (legally, of course) to explain or demonstrate a product makes advertising cheap and easy.  

With a variety of tools available, it’s never been easier to start making deepfakes without needing extensive technical knowledge. Here’s a closer look at this process.

How Easy is it to Create Deepfakes?

At the heart of deepfake technology lies a complex system known as Generative Adversarial Networks (GANs). GANs consist of two neural networks—the generator and the discriminator—that work against each other to produce realistic images or videos.

While the underlying technology is sophisticated and requires a deep understanding of machine learning to develop from scratch, most consumer-level products can simplify this process. Users can leverage these advanced algorithms without needing to grasp the technical intricacies, making deepfake creation accessible to the general public.

Examples of Deepfakes

Here are some recent examples of deepfakes that have fooled the public and took off in popularity.

Deepfaked Back to the Future: A popular deepfake video created by YouTuber EZRyderX47 replaced the characters Marty McFly and Doc Brown with Tom Holland and Robert Downey Jr., respectively. The video gained significant attention, amassing over 11 million views due to its creative execution.

Taylor Swift Deepfake Scam: In January 2024, a deepfake video featuring a fake Taylor Swift promoting a Le Creuset cookware giveaway circulated online. This deceptive video tricked some fans into believing it was real, leading to financial scams.

Deepfake of Morgan Freeman: A highly convincing deepfake of Morgan Freeman was shared on social media, raising concerns about the potential use of such technology in political contexts, especially during elections.

Tom Cruise Deepfake on TikTok: A TikTok account dedicated to deepfake videos of Tom Cruise gained popularity for its humorous and realistic portrayals of the actor engaging in everyday activities, showcasing the technology's entertainment potential.

Deepfake of Richard Nixon: A deepfake video featuring a digitally altered Richard Nixon announcing a failed moon landing was created using actual footage from his resignation speech. This example illustrates how deepfakes can manipulate historical narratives.

How to Spot a Deepfake: Tips and Techniques

Deepfake technology has advanced significantly in recent years, making it increasingly challenging to distinguish real videos from manipulated ones. However, by knowing what to look for, you can sharpen your ability to identify potential deepfakes. Here are some key signs and techniques:

Visual Cues

1. Facial Features: Scrutinize the face closely. Look for any inconsistencies or oddities in skin texture, tone, and facial expressions. Is the skin too smooth or wrinkled? Do the facial features seem proportional and properly aligned?
2. Eyes and Eyebrows: Pay special attention to the eyes and eyebrows. Unnatural movements, lack of detail, or strange reflections in the eyes can be red flags. Deepfakes often struggle to replicate realistic eye movements and blinking patterns.
3. Mouth and Speech: If the video shows someone speaking, carefully observe the synchronization between lip movements and the audio. Mismatched or unnatural speech patterns can indicate a deepfake.
4. Lighting and Shadows: Be on the lookout for inconsistencies in lighting and shadows. Deepfakes may have difficulty accurately rendering how light interacts with faces and surroundings, resulting in strange shadows or highlights.
5. Background Details: Don't just focus on the subject - examine the background too. If the person is crisp and clear while the background appears blurry or poorly rendered, it could suggest manipulation. Look for any visual artifacts or glitches.

Behavioral Cues

1. Expressions and Emotions: Watch for facial expressions or emotions that seem exaggerated, robotic, or out of character for the individual. Deepfakes often fail to capture the subtle nuances of authentic human expressions.
2. Mannerisms and Gestures: Consider the person's unique mannerisms, gestures, and body language. Do they match what you know about this individual? Any inconsistencies can be warning signs.
3. Unnatural Details: Keep an eye out for small details that seem artificial or out of place, such as strange facial hair, misaligned moles, or teeth that don't look quite right. These discrepancies can indicate a deepfake.

Contextual Factors

1. Plausibility and Probability: Think critically about the content and context of the video. If it portrays a public figure doing or saying something that seems exaggerated, unrealistic, or wildly out of character, be very skeptical.
2. Source and Platform: Always consider the video's source. Exercise extra caution with content from unverified, dubious or unfamiliar platforms. Reputable news outlets are less likely to distribute manipulated media without thorough verification.
3. Cross-Referencing: When in doubt, cross-reference the video with other trusted sources. See if the same claims, quotes or events are being reported elsewhere from reliable outlets. A lack of corroboration could suggest a deepfake.

Expert Analysis: In some high-stakes situations (legal cases, election integrity issues, etc.), consulting digital forensics experts who specialize in analyzing manipulated media can provide more definitive answers.

How to Create Your Own Deepfakes

Here’s the process for creating your own deepfake, using one of the top tools on the market, AKOOL, as an example.

Step-by-Step Guide: How to Start Making a Deepfake

Let’s go through the basic steps for all three deepfake methods.

Changing the Face in a Photo

AKOOL’s Tools menu is where you’ll find all features. 

Choose Tools / Face Swap.

‍

Find the photo whose face you want to swap. Drag the image file into the box on the right, or click Choose Files to locate it.

AKOOL analyzes the image, and identifies the detected face.

The face used for swapping must come from the Choose Face list on the right. Use the Add icon to bring in a photo, or use one of the provided defaults. For the photo you bring in, AKOOL will run the same face detection.

Click that face to add it to the Detected Faces list.

Face Swap also has a Re-age slider, to add or subtract years from the final face, and a Face Enhance option, for removing blemishes and the like.

All AKOOL tools have an easy-to-find “run” button at the lower right. Click High Quality Face Swap.

Another smart AKOOL UI feature: all generated results are stored in one place. Click Results at the top right.

The Swap Results list holds the final product- the new face added to the original photo. Click Download to save it.

Changing a face in a video

For the next deepfake feature, record or download a video with a clear view of one or more faces. This is the video used in this example. 

The process is the same as a single face swapping image. In the Face Swap tool, just bring in the video instead of a photo. The Detected Face list produces not just one face, but several face views, according to the various angles identified in the video. 

You can bring in a single face view from the Choose Face list. But for even better results, use photos of the swap face that match the angles found in the video.

As before, click High Quality Face Swap, and find the video in Swap Results.

It’s pretty amazing to see the new face replacing the original in every angle.

Creating a Talking Photo

Our final deepfake is likely familiar to you, if you pay attention to AI developments. 

Maybe you’ve seen videos in which the subject of a famous work of art (think Mona Lisa) speaks. Talking paintings and cartoons don’t fall into the deepfake category - everyone knows they’re not real. But reality gets a bit more shaky when a photographed face is doing the talking. And it’s shockingly easy to do.

In AKOOL, the tool this time is Talking Photo. Just upload the photo into the UI.

Enter your script in the Text box, inserting line breaks for pauses. Choose a male or female voice.

Choose the speaker from the list of names. Each name has a play button you can use to hear the voice.

To hear the script delivered in the selected voice, click the large play button. You can adjust speed if needed.

Scroll down and click High Quality Generate.

Then open your results and check out your deepfake. 

Potential Applications and Impact

While there are obvious benefits of AI video face swap tools, we can’t argue that there is no downside. 

Think of faked videos of politicians taking bribes or purchasing drugs. Or celebrities swapped into pornographic content.

Beyond these obvious examples, there are also more mundane use cases involving ordinary people, with far-reaching implications.

Deepfake cloned voices have been used in a variety of crimes and scams, such as mimicking a loved one’s voice in distress to extract ransom. Deepfake social media ads have used celebrities or other trusted figures to convince viewers to take some sort of financial action. 

And while deepfakes used to be rather easy to spot, they will only grow in sophistication. AI deepfake analyzers must keep pace with the deepfake tech itself.

But despite the drawbacks, many experts feel that deepfake benefits outweigh the risks. Here are a few positive examples:

• News organizations can use real or custom news presenter avatars, capable of presenting current events in multiple languages to audiences around the world. without the need for studios and AV equipment.
• Families can use deepfake technology to “recreate” elderly or deceased loved ones, so that young children can see what their great-grandparents looked and sounded like. 
• Museums and educational facilities are using deepfakes to produce interactive, “living” historical figures. As one example, a Salvador Dali deepfake was created to enable museum-goers to listen to the famous artist in his own words. Hours of video interviews were processed and analyzed by AI, in order to produce the startling real looking Dali, speaking in his unique accent. 
• When celebrities are willing participants, they can lend their famous faces and voices to good causes. In one such case, David Beckham partnered with Malaria no More, to produce public service announcements in nine languages.

Get Started Now

To make your own deepfakes, check out AKOOL for yourself. The free account comes with enough starter credits to generate several examples.

But always be aware of legal and ethical considerations during your deepfake journey. Fake responsibly!‍