AI Video Company Synthesia Launches Express-3 Digital Human Model
AI video company Synthesia has launched the next-generation digital human model, Express-3, claiming it to be the most advanced avatar model to date: it performs with perceivable script emotions, has more natural lip-syncing and body language, and generates content approximately twice as fast. This model also drives Style Avatars, allowing users to create stylized characters in the Avatar Builder using prompts or presets, and is available for all packages.
The release notes specify that Express-3 will replace Express-2 on the platform, while Express-1 avatars remain unaffected. Emotional performances can switch between contexts such as happy, serious, neutral, and friendly; visuals are clearer, lip-syncing aligns better with speech, gestures are more script-aligned, and gaze issues are reduced. The company also introduced a text assistant that completes planning, brand alignment, and revisions in a single conversation, consolidating workflows into one chat box.
Synthesia targets corporate training and marketing short videos, where buyers seek repeatable, translatable on-screen personas rather than cinematic long shots. In evaluation metrics, short clips are easier to pass blind tests for "human-ness," while those over a few minutes may still reveal repetitive eye movements and micro-expressions. Style Avatars expand the realistic digital human product line to include cartoon and brand mascots, reducing the need for custom shoots.
The generation speed doubling directly impacts the queuing and per-clip billing experience. The availability of all packages means that free or entry-level users can also access the new rendering, aimed at retention rather than just providing enterprise seats. The interactive digital human route is still separately planned for listening and viewing, with Express-3 initially addressing one-way content generation.
Market-wise, this represents a generational shift in corporate video generation models. Buyers are companies looking to produce training videos in bulk; sellers are vendors who maintain platform default rendering through emotion and speed. Funds flow from subscription seats to rendering queues. Beneficiaries are clients who have integrated brand digital humans into templates; those under pressure are competitors relying on old lip-syncing methods and small studios shooting live-action or digital humans on a per-clip basis. The event is driven by product releases.
On a supplementary note, "approximately twice as fast" refers to the official generation delay metric, excluding script and review time. Copyright and portrait rules for Style Avatars depend on the generation terms.
Source: Public Information
ABAB AI Insight
The bottleneck in corporate video has never been the shots, but the availability of on-screen talent. Synthesia has turned availability into a model version number: Express-3 replaces Express-2, and switching all packages means a forced upgrade of the default avatar. Emotion perception translates script emotions into expression tracks, while lip-syncing and gestures aim to minimize pauses during quality checks. Doubling the speed allows the same seat to produce more clips in a day, diluting subscription costs.
The capital path is "render engine locking workflow." The assistant consolidates planning and final output into chat, while the Avatar Builder turns stylized characters into self-service generation. The more clients store their brand faces on the platform, the higher the migration costs. Money flows from annual fees into GPU queues. The motivation is to retain enterprise workspaces using default model quality when competitors like HeyGen compete on price.
In comparison to the iteration of digital voice tools and the translation dubbing that spreads the same face across multiple languages: both convert the fixed costs of live-action shooting into subscriptions. The industry phase is shifting from talking avatars to those that can act according to emotions, with control over whose rendering becomes the template default. Long videos still reveal flaws, but short training clips are already sufficient.
Structural changes belong to technological substitution. The mechanism is: once on-screen talent can be version upgraded, live-action shooting retains only strong emotions and long shots; pricing for corporate content shifts from producer days to seat and queuing duration. Therefore, the model upgrade is not just about special effects updates, but about removing human resources from in front of the camera.
ABAB News · Cognitive Laws
Corporate videos first eliminate availability, then eliminate shots
Once default rendering is upgraded, it becomes harder to move workspaces
When emotions can be written as tracks, voiceovers no longer need actors present