Morph 1.0
The same signing, a new face. Morph 1.0 re-renders a signing video as a different person and keeps its timing, frame count and size exactly.
Read the launch postTalksign Morph
Our video-to-video model. Morph re-renders any signing video as a new person, from just one photo.
Announcements
Morph 2.0
Our next video-to-video model, fine-tuned on sign language with Echo 2.0. Finger accuracy on fast signing is the first thing it improves.
Join the waitlistAvailability
Morph 1.0 powers Sign Morph in Talksign Canvas: upload a signing video, choose a person, download. It runs on the same model as Echo 2.0.
It takes a signing video of any length, frame rate or size, plus one photo of the new person; a neutral, front-facing photo works best. Morph redraws every frame, following the hands, body and face, and returns a photorealistic video at up to 720p with the same frames, timing and size, plus a hand report for review. It has been tested on American Sign Language (ASL) signing.
Processing takes longer than the video itself, so Morph is for produced content, not live calls.
To get started, request API access or email dev@talksign.co.
Use cases
Sign language video has always been tied to the person who recorded it. Morph keeps the signing and changes the person, so teams can publish produced sign language content in new ways:
01
Media and entertainment
Re-present signed episodes, trailers and social clips with one presenter, without booking the studio again.
02
Brands and marketing
Give your brand one signing presenter across campaigns and product videos, even when different people signed them on different days.
03
Education
Publish a whole course with one familiar presenter, so learners follow the same face from lesson to lesson.
04
Public services
Share signed announcements and guidance with an official presenter, while the staff who signed them keep their faces private.
05
Healthcare
Produce signed patient information with a consistent presenter, and let the signers who recorded it stay anonymous.
06
Creators and community
Deaf creators sign as a character or avatar, with their own signing, timing and expression intact. The signing still belongs to them.
Benchmarks
Morph 1.0 keeps the new person consistent, with the highest identity similarity of the systems below, and moves smoothly from frame to frame.
Frame continuity against research systems
| Morph 1.0Talksign | SignMimicResearch | Animate AnyoneResearch | MimicMotionResearch | Magic PoseResearch | |
|---|---|---|---|---|---|
| SSIM-TShigher is better | 0.908 | 0.963 | 0.934 | 0.904 | 0.898 |
| PSNR-TShigher is better | 26.88 | 28.90 | 27.49 | 23.75 | 25.24 |
Identity similarity against avatar systems
| Morph 1.0Talksign | HeyGenAvatar system | KlingAvatar 2.0Avatar system | OmniHuman-1.5Avatar system | |
|---|---|---|---|---|
| IDC, DINOv2higher is better | 0.80 | 0.79 | 0.77 | 0.71 |
Known limitation. Finger detail on fast, continuous signing is our open problem: none of our four test videos meets our hand-accuracy bar yet, and every output comes with a hand report that flags frames for review. It’s the first thing Morph 2.0 improves. See the figures in the model card.
Other systems’ values are from their papers: frame continuity from SignMimic (ASL 50K), identity from JoyStreamer. They were measured on different test sets from ours, so they show the level each system reaches, not a head-to-head. Morph 1.0: 5 renders of 4 test videos, 1,185 frames. Details are in the model card.
Trust and safety
Morph may only be used with the consent of both the signer in the video and the person in the photo. Outputs are machine-generated and should be labelled as such, and impersonation and deceptive use are not allowed. Re-rendering changes appearance, not authorship: the signing stays the original signer’s work. We’re preparing a blind study where fluent Deaf ASL users judge whether meaning, clarity and naturalness are kept, and we’ll publish the results.
Frequently asked questions
When should I use Morph?
Use Morph for produced sign language content: a consistent presenter, a creator’s character, or a signer who wants to keep their face private. It isn’t for live calls, and for legal, medical or emergency communication the original signer or a qualified interpreter should be used.
How long does it take?
Longer than the video itself, so Morph is for produced content you can prepare ahead, not live calls.
What photo works best?
A neutral, front-facing photo. The photo’s pose and smile can carry into the first frames.
