Mechanika synchronizacji ruchu warg
Reading phonemes, building visemes, and judging lipsync
Treat speech as a modular system. Learn the mechanical requirements of each sound, how to build lipsync shapes from FACS, when to split your shapes or go past FACS, and why.
For riggers, AI teams, and animators who need to strategize and evaluate lipsync. Designed for human-made or generative systems. American English is the main example but principles apply to any major language. For more multilingual help, add the guided Multilingual Survival Kit.
Taken by artists at Blizzard, XBOX, and Larian · One 3.5-hour live session · 3-year recording access
primary shape pomadka do ust optional podbródek
Visemes aren’t set in stone.
Speech is a system of moving parts. Shape needs change with context and technical capabilities. Fixed viseme charts and off-the-shelf sets neglect the nuance.
Viseme groupings are contingent on context and capabilities.
if rig has no tongue → grupa th z kg
if smiling during pbm → reduce smile or allow small gap
if energy = low → reduce articulation
if fidelity = low–medium → grupa l oraz n
if fidelity = high → split dark and light l
Don’t
- Copy a viseme set built for someone else’s rig and use case
- Split sounds your fidelity doesn’t need, or merge ones it does
- Allow AI to hyperarticulate every sound
Do
- Build your viseme set for your rig and level of detail
- Match articulation to emotion, energy, and word context
- Learn the key properties of each sound and the why behind your decisions
How the system works
Built from phonetics, motor-control research, FACS, and years of studying real speech for face tracking and AI dubbing evaluation.
1 · /f/ = f, ph, gh
Demystify the IPA
How a phoneme chart is organized and what it can tell you about the face.
2 · p/b/m: lips meet
Know what each sound needs
Requirements for each phoneme ranked from strict to flexible, so you know what to bend without breaking.
3 · pucker + cincher
Build it from parts
The main drivers, the supporting shapes, and the contexts that sustain or strain them.
4 · smile + /m/
Let context bend it
Neighboring sounds, emotion, and energy influence shape outcome. How to handle these variables.
5 · no tongue?
Fit it to your rig
Gain the tools to build a modular viseme set customized for your rig and fidelity.
There’s no universal viseme set. Answer a few questions and build the one your project needs.
Does your rig have a tongue? no → Merge shapes that differ primarily by the tongue, like th oraz kg
yes → How much detail do you need? low → Merge l oraz n high → Split light l and dark l
The full decision tree is one of the workshop handouts.
Sample lecture slides
Reference slides from the workshop. Click for full size.
How to build widened vowel shapes
Visible constraints by articulator
How to build crossed form pbm shapes
Live session
Saturday, March 13, 2027 @ 9am PT (12pm ET · 5pm UK · 6pm CET).
Session length is ~3.5 hours. Includes: two lecture blocks, a short break, and 30 minutes of Q&A.
Can’t make it live? No problem. Everyone who signs up receives 3 years of recording access and copies of the PDF slides to keep.
~9:00–10:20
Part 1: Foundations, Priorities, and Strategies
Sound vs. spelling and how to read a phoneme chart. Consonants: strong vs. weak, requirements, and how to reason about your own groupings.
10 min
Break
~10:30–11:50
Part 2: Shape Building and Modular Systems
Vowels and diphthongs, shape formulas, critical FACS combinations, emotion and energy, relative contrast, and a real-footage case study.
30 min
Live Q&A
Optional add-on · $92
Multilingual Survival Kit
How to use linguistic foundations to approach and evaluate languages you don’t speak.
- Step-by-step guide to assessing lipsync in languages you don’t speak
- How to leverage IPA charts
- Examples from Portuguese, French, and Mandarin
- The phonology references and strategies I use when there is no clear path
Every ticket includes Viseme reference card Lipsync evaluation checklist Viseme set decision tree 3 years of recording access PDF slides
Who it’s for
Riggers and TDs
Build a viseme set with custom phoneme mappings. Know which shapes are essential, helpful, or antagonistic to the mechanical needs of each sound.
AI and ML teams
Establish criteria for judging generated lipsync, a shared vocabulary for failure modes, and viseme targets for data specs.
Animators
Learn how to prioritize shapes and disentangle articulation needs from emotional expressions.
Some FACS familiarity helps. New to FACS? Start with my free resources, or email me to receive a 10% course combo discount for Lipsync Mechanics + FACS Cram Session.
Dates and rates
Save up to $315 with the Super Early Bird rate. Add the Multilingual Survival Kit to any ticket for $92.
Now · Super early bird
$380
Through Dec 22, 2026
Early bird
$440
Through Jan 22, 2027
Standard
$520
Through Feb 22, 2027
Late
$695
Until Mar 12, 2027
For studios
Train your whole team in a private session, on your schedule.
Private studio workshop
Book a team session with a private Q&A. Choose date and time. From $4,200.
Add-on: lipsync audit
I review your viseme set, rig, or model output against the workshop’s requirements and send written notes. Contact me for pricing.
Studios pay by invoice. Email facetheFACS@melindaozel.com with your team size.
Your instructor
Melinda Ozel is a facial systems expert specializing in FACS, facial anatomy, and lipsync. Her work spans face tracking, film, and robotics, from scaling face-tracking data strategy at Oculus to consulting on AI film technology and humanoid robots.
- Obserwuj niebo: AI-based lipsync expert (film, credited)
- Oculus LipSync: expression researcher
- Meta Quest Pro face tracking: expression researcher
- Co-inventor on three avatar expression patents
- Tutaj: AI facial expression consultant with Metaphysic
“It’s a very focused course that is useful immediately, no matter what level your knowledge about lipsync is.”
Stijn Valkenborg, Senior Technical Animator, Larian Studios
Pytania
Where is the course held?
Zoom. You will receive an invite once you sign up.
What if I can’t attend live?
There is no requirement to attend live. Everyone who signs up gets 3 years of access to the lecture recording, a 320+ page PDF slide deck, a viseme reference card, a lipsync evaluation checklist, and a viseme set decision tree.
Is this still relevant to me if I don’t use FACS?
Yes, you can map your controls to analogous shape systems or construct equivalent joint combinations. The formulas describe what each shape has to do, not a specific rig’s sliders.
Which languages does it cover?
American English is the worked example. The Multilingual Survival Kit gives you the tools to apply learned concepts to any major language, even ones you don’t speak.
We use AI lipsync. Is this still useful?
Yes. The requirements work as criteria for judging generated output, writing data specs, and explaining where and how a model failed or succeeded.
Saturday, March 13. Save your seat.
9am PT · online · 3 years of recording access