Synthesia is an AI video platform used by enterprise teams to produce training, sales, marketing, and internal comms videos without cameras, actors, or studios. This launch introduces Express-3, the company's newest avatar model, which powers the AI presenters that speak a user's script inside the Synthesia editor. The new avatars follow the sentiment of the script, so they can be happy, serious, neutral or friendly depending on what has been written, and Express-3 introduces more script-aligned gestures and movement so avatars feel less static and more human.
The upgrade targets the details that give away synthetic video: sharper faces, tighter lip sync, more varied hand and body gestures, and fewer eye-gaze glitches. Videos using Express-3 can also generate up to 2x faster, reducing the wait between editing and reviewing , and the model powers new creation workflows including new stock Avatars, Style Avatars, and Avatar-to-Video through Assistant, having been developed and evaluated through internal head-to-head testing focused on lip sync, body motion, emotional control and overall publishability. For teams already producing at volume on Synthesia, the practical wins are shorter render cycles and less manual re-shooting of scenes where an avatar's tone did not match the copy.
The launch was posted by CEO Victor Riparbelli, who co-founded Synthesia in 2017 alongside Prof. Matthias Niessner, Prof. Lourdes Agapito and Steffen Tjerrild. Express-3 lands as part of the broader Synthesia 3.0 push toward avatars that can listen and respond as well as speak, and Synthesia Assistant and the full range of Express-3 Avatars are available now for all users on all plans.
Comments (9)
Sign in to join the discussion.
P
Priya K.Sep 10, 2026
Express-3 sounds great but the real test is whether my avatar stops doing that weird half-blink between sentences. That thing haunts my compliance training videos.
T
Tomas LindqvistSep 10, 2026
The cut at 0:12 where the avatar shifts from cheerful to serious mid-sentence is genuinely the best part of the video. Whoever edited this understood the assignment.
M
Mika KurodaSep 10, 2026
Every avatar demo cherry-picks the three seconds where lip sync works. Show me a 90 second unbroken take or I'm not buying it.
O
Ovie A.Sep 10, 2026
Sentiment-aware avatars are cool until my quarterly earnings script gets read like a breakup letter. Please tell me there's a manual override.
R
rafiSep 10, 2026
Tweet copy is doing that bullet-list-with-dashes thing every AI launch uses now. At some point the format itself becomes the giveaway.
Z
Zara AbbasiSep 10, 2026
Before we get excited, where is the training data residency documented and can enterprise customers get contractual guarantees the likeness isn't reused? Asking because legal will.
K
kojiSep 10, 2026
Curious what backbone this is running on and whether context adaptation is a separate module or baked into the diffusion step. Any paper or tech note dropping?
L
Leo MarchettiSep 10, 2026
One more thing for the roadmap: let me clone my avatar's gestures from a 30 second reference clip of me actually presenting. Sentiment inference is nice but I want my hand-waving back.
D
Dana H.Sep 10, 2026
You already had me at Express-2. The fact that you're shipping a v3 while competitors are still writing landing pages is the actual flex here.
Work With Us
Book an intro call to explore our suite of services. We might have a waitlist.
Synthesia
All-in-one AI Video platform for business
All-in-one AI Video platform for business
Tags
TLVC Rating
Community Rating
No ratings yet
to rate this launch.
Synthesia is an AI video platform used by enterprise teams to produce training, sales, marketing, and internal comms videos without cameras, actors, or studios. This launch introduces Express-3, the company's newest avatar model, which powers the AI presenters that speak a user's script inside the Synthesia editor. The new avatars follow the sentiment of the script, so they can be happy, serious, neutral or friendly depending on what has been written, and Express-3 introduces more script-aligned gestures and movement so avatars feel less static and more human. The upgrade targets the details that give away synthetic video: sharper faces, tighter lip sync, more varied hand and body gestures, and fewer eye-gaze glitches. Videos using Express-3 can also generate up to 2x faster, reducing the wait between editing and reviewing , and the model powers new creation workflows including new stock Avatars, Style Avatars, and Avatar-to-Video through Assistant, having been developed and evaluated through internal head-to-head testing focused on lip sync, body motion, emotional control and overall publishability. For teams already producing at volume on Synthesia, the practical wins are shorter render cycles and less manual re-shooting of scenes where an avatar's tone did not match the copy. The launch was posted by CEO Victor Riparbelli, who co-founded Synthesia in 2017 alongside Prof. Matthias Niessner, Prof. Lourdes Agapito and Steffen Tjerrild. Express-3 lands as part of the broader Synthesia 3.0 push toward avatars that can listen and respond as well as speak, and Synthesia Assistant and the full range of Express-3 Avatars are available now for all users on all plans.
Comments (9)
Express-3 sounds great but the real test is whether my avatar stops doing that weird half-blink between sentences. That thing haunts my compliance training videos.
The cut at 0:12 where the avatar shifts from cheerful to serious mid-sentence is genuinely the best part of the video. Whoever edited this understood the assignment.
Every avatar demo cherry-picks the three seconds where lip sync works. Show me a 90 second unbroken take or I'm not buying it.
Sentiment-aware avatars are cool until my quarterly earnings script gets read like a breakup letter. Please tell me there's a manual override.
Tweet copy is doing that bullet-list-with-dashes thing every AI launch uses now. At some point the format itself becomes the giveaway.
Before we get excited, where is the training data residency documented and can enterprise customers get contractual guarantees the likeness isn't reused? Asking because legal will.
Curious what backbone this is running on and whether context adaptation is a separate module or baked into the diffusion step. Any paper or tech note dropping?
One more thing for the roadmap: let me clone my avatar's gestures from a 30 second reference clip of me actually presenting. Sentiment inference is nice but I want my hand-waving back.
You already had me at Express-2. The fact that you're shipping a v3 while competitors are still writing landing pages is the actual flex here.