Over the past five years, AI face swapping capabilities have advanced at a staggering rate. What began as relatively primitive celebrity selfie filters have fast transformed into unnervingly realistic video manipulation tools now used everywhere from Hollywood blockbusters to political smear campaigns.
To illustrate just how quickly this field has progressed:
-
In 2017, an AI lab produced the first end-to-end face swapping algorithm for believable results framed around consent. Within a year, commercial apps like Zao with far more nefarious intents drew hundreds of millions of downloads across China.
-
Investment in synthetic media startups has ballooned from less than $50 million in 2018 to over $1.5 billion in anticipated 2023 funding. Uses now span entertainment, politics, adult content, education and more.
Figure 1 – Venture funding for AI-powered face/voice swapping startups has risen over 30x since 2018. Source: Pitchbook
And an array of new consumer tools continue pushing the boundaries of edits possible on images, videos, even live streams:
-
Want to sport Brad Pitt’s photorealistic visage while live streaming games as an anime character? Now possible for a few hundred bucks.
-
Prefer putting Nicolas Cage’s face on actress Florence Pugh in the latest Marvel trailer? Freely accomplishable in seconds with today’s chops.
-
Or imagine dropping Kanye West’s viral vocals over Bruno Mars performing live at the Super Bowl filtered through three layers of Disney cartoon princess translucencies… What a time to be alive!
But clearly such limitless alteration capacity poses many ethical risks if left unchecked. So in this guide, we‘ll explore the current landscape of AI face swapping—from playful selfie filters to professional film editing suites—looking carefully at which tools balance creative possibility with responsible practices.
Surveying the Current State of Face Swap Technology
Before diving into the tools, let‘s briefly level-set on what face swapping actually entails.
The phrase “face swap” refers to using AI algorithms to seamlessly replace or alter faces in digital photos, videos and even live streams. Computer vision techniques enable mapping things like facial contours, skin tone, expressions and lighting from a source face onto a target. The latest methods produce disturbingly convincing results—hence the rise of the term “deepfakes”.
Figure 2: Convolutional neural networks power most state-of-the-art face swapping models by pinpointing spatial relationships between facial features. Source: Deeptraque
What unique strengths and weaknesses distinguish the expanding landscape of providers enabling face swapping edits?
Below we explore 13 top contending tools based on factors like:
-
Technological Maturity: How convincing are the face swaps produced? Do results translate seamlessly to profile views and video? How resilient are models against artifacts and glitches?
-
Ease-of-Use: Does the interface cater more to casual smartphone users or technical experts? How much configuration vs point-and-shoot simplicity is involved?
-
Customization Range: Can users fine tune parameters like skin tone matching, smoothing, image backgrounds? Does it offer filters and stylization options?
-
Ethics & Consent: Have developers implemented safeguards around potential misuse? How are personal photos and data handled? Do outputs disclose edits made?
Now let’s see how today‘s top offerings compare…
13 Leading Face Swap Tools Stacked Up
Reface
- Launched: 2017
- Developer: Reface (Ukraine)
- $199M raised to date across 5 rounds
- 100M+ downloads fuels rapid iteration speed
Founded by four Ukrainian engineers, Reface produces incredibly fluid video face swaps fine-tuned using a custom GAN architecture. Their models smoothly track expressions, angles and lighting even on mobile devices.
However, the app’s terms of service absolve liability for editing created without consent. And exported clips embed no signals identifying manipulation.
Figure 3 – Reface Face Swaps Convince Even In Profile Video. Source: Reface
Wombo
- Launched: 2021
- Developer: Wombo Labs
- 2M downloads within months of launch
- Viral sensation for musical lip-sync clips
This fledgling startup explode onto the scene for its musical lip sync face swap clips—no user media required. Just pick a track and watch a fantastical figure or meme template dance across the screen.
While outputs embed an identifiable Wombo logo, the app focuses more on virality than realistic video quality. But its instant hilarity factory makes Wombo tough to compete with for sheer entertainment.
Figure 4 – Wombo‘s Musical Memes Use Fun Face Swaps. Source: Wombo
Zao
- Launched: 2019
- Developer: Momo (China)
- 200M downloads shortly after launch
- Leverages Momo’s huge database for training
Upon release, face swapping app Zao immediately triggered privacy concerns given its creator Momo’s access to a database of millions of users‘ photos and videos. Training on such data resulted in incredibly resilient swap quality.
Zao also endured some early criticism around female face support and closely resembled eastern appearance norms. But rapid improvements have since addressed complaints.
Figure 5 – Zao Can Swap Faces From Extreme Perspectives. Source: Zao App
Snapchat Filters
- Launched Early Lenses in 2015
- Developer: Snap Inc.
- 265M daily active users fuel ML filter innovation
As pioneers of mobile augmented reality face effects, Snapchat’s machine learning algorithms have access to more facial data than nearly any platform on earth. This allows their filters to deform features fluidly while anchored around key positional nodes.
But Snapchat caters filters more for friendly fun than convincing fraud. Creations lock tightly within Snap’s vertically integrated ecosystem as the company shifts focus towards AR smart glasses.
Figure 6 – Snapchat Filters Emphasize Amusing Transformations Over Realism. Source: Snapchat
TikTok Effects
As home to viral meme proliferation, TikTok makes an obvious vessel for next-generation face swapping capacity through its augmented reality Effect tooling.
Accessible drag-and-drop interfaces encourage everyday users to develop effects applied using the smartphone’s front-facing camera. Considering TikTok’s breadth of facial data, their algorithms could produce incredibly realistic results.
However for now, effects remain focused on facial ornamentation over replacement—likely to maintain safeguards against harmful deepfakes during such impressionable growth. Quality face swaps may still emerge in time as creators experiment with what captivates audiences.
Figure 7 – Most Trending TikTok Effects Playfully Decorate Faces Rather Than Swapping Them. Source: TikTok
FaceApp
- Launched: 2017
- Developer: Wireless Lab (Russia)
- 200M+ downloads to date
- One of the earliest viral photo editors
This pioneer face editing app took off around 2017 for its dramatic aging filters morphing users into elderly versions of themselves. But amidst concerns about photos getting stored indefinitely on external servers, criticism around their handling of personal data soon emerged.
And FaceApp integrates few face swapping capabilities natively so far. However viral filter concepts stemming from their aging effects offer valuable petitions for face swap tool creators aiming to captivate audiences.
Figure 8 – FaceApp‘s Signature Aging Filter Dramatically Alters Faces Without Swapping Them Outright. Source: FaceApp
Fotor
Yet another photo enhancement suite joining the face swap movement…
Fotor offers a polished touch-up workflow combined with batch automated face swapping tools useful for advertising campaigns and collateral needing fresh models. The software runs natively across web, mobile and desktop apps.
Outputs showcase solid realism but still embed slightly visible artifacts around contours in side perspectives. Their integration posture also leaves opportunity for Canada-based Picmunk to own the prosumer editor space.
Figure 9 – Fotor Allows Batch Face Swapping To Quickly Vary Models Across Campaign Images. Source: Fotor
B612
As pioneers around mobile beauty effects, B612 pivoted their augmented reality features more expressly into facial overlays and swapping focused on social sharing.
While so far limited to the B612 app ecosystem itself, their face tracking and skin smoothing algorithms evolved through millions of Asian selfies could soon pose formidable competition integrated into other video editing toolchains.
Figure 10 – B612 Focuses More On Decorating/Altering Faces Rather Than Swapping Them Out. Source: B612
FaceSwap Live
Representing iOS-focused entries, Face Swap Live performs surprisingly responsive face swaps right within the live camera feed. This grants a window into just how quickly machine learning algorithms can infer facial coordinates and relight new faces anchored to the scene.
Files get saved natively to the Camera Roll rather than externally too. But the app remains constrained to still images so far. Once expanded to video, Face Swap Live may grow into a formidable mobile editing contender.
Figure 11 – Face Swap Live Can Track And Rotate Faces In Real-Time During Recording. Source: FaceSwap Live
Avatarify
A fledgling face swapping company from machine learning experts, Avatarify aims to expand possibilities for video game streamers and content creators through augmented reality avatar effects applied in real time using webcam footage.
This young startup balances realistic face swapping with a strong emphasis on transparency and consent absent from several leading alternatives. It will be exciting to see how Avatarify stimulates creative applications while progressing responsibly.
Figure 12 – Avatarify‘s Advanced Face Tracking Will Soon Power Webcam Avatar Overlays. Source: Avatarify
Lyrebird
Shifting into the audio realm, Lyrebird spun out from academic voice mimicry research to commercialize speech synthesis models achieving unprecedented realism.
By analyzing as little as one minute of sample audio, Lyrebird algorithms generate strikingly convincing vocal imitations packaging familiar cadence and delivery—an obvious avenue for exploiting consent. Hence critics urge caution around societal impact as capabilities scale up.
Figure 13 – Lyrebird Mimics Nuanced Features of Someone‘s Voice From Small Samples. Source: Lyrebird
Synthesia
Founded in 2017, Synthesia produces stunningly photoreal AI avatars mimicking a person‘s likeness, voice and mannerisms using just a bit of video footage.
But unlike many startups rushing services to market, Synthesia targets only approved enterprise use cases until models mature further. Their caution acknowledges risks of exploitation under rapid amplification—a framing sorely lacking in Synthesia’s less scrupulous contemporaries.
Figure 14 – Synthesia Avatars Capture Subtle Mouth Articulations Using Retargeted Video. Source: Synthesia
D-ID
Rather than swapping faces outright, this company focuses on anonymizing images and video to generate synthetic data free of individually identifiable linkages to train machine learning models.
D-ID offers services surrounding sensitive applications like healthcare where personal info poses barriers to gathering data at scale. They also consult on securely documenting evidence of war crimes through identity cloaking.
The company’s toolkit increasingly blends synthesized faces and voices toopens possibilities for VR therapy, disease research and beyond without violating patient privacy. AI ethics groups applaud the consideration and care of D-ID’s approach.
Figure 15: D-ID leverages AI generate anonymized facial data for training machine learning models without privacy violations. Source: D-ID
