HeyGen Avatar 6 tested: lip sync, Turbo vs regular and OmniHuman 2
I tested HeyGen Avatar 6 on fal the day it launched: lip sync 8/10, Turbo 7.5/10 at half the price, and a head-to-head with OmniHuman 2.

👁 118 reads
HeyGen Avatar 6 arrived on fal.ai on 17 November 2026 in three modes - from a photo, from a reference video of an illustrated character and from a stock HeyGen avatar - each in a regular and a cheaper Turbo version. I tested it the same evening with a fictional woman who does not exist and an illustrated paper-craft mouse, then put it side by side with OmniHuman 2 and MiniMax H3 Max Lip Sync on the same photo and the same voice. Short version: the lip sync is very good, Turbo costs half and is only slightly worse.
Avatar 6 in short
In my test Avatar 6 scored 8/10 for lip sync and Turbo 7.5/10. The lips round correctly on hard consonants and close in longer pauses. In very short pauses the character smiles with visible teeth instead of closing the mouth - the only giveaway. On fal one minute of 1080p video costs USD 6 (regular) or USD 3 (Turbo).
What exactly launched
HeyGen is the company behind one of the best-known talking-avatar apps. Avatar 6 is its new engine, and now it can also be called through the fal API, paying per second of finished video with no subscription.
| Mode on fal | What you provide | Regular | Turbo |
|---|---|---|---|
| Image to Video | a photo of a person + a speech recording (optional motion prompt) | USD 0.10/s | USD 0.05/s |
| Reference Video to Video | a short video of an illustrated or animated character + a speech recording (optional photo of it) | USD 0.10/s | USD 0.05/s |
| Identity to Video | a stock HeyGen avatar + a text script with a HeyGen voice, or your own recording | USD 0.10/s | USD 0.05/s |
Prices are for 1080p. In 2K it is USD 0.145/s (regular) and USD 0.0875/s (Turbo). The recording sets the length of the video, and Identity to Video outputs 16:9, 9:16, 1:1, 4:5 or 5:4. One important limit: the reference-video mode does not accept videos of real people - only illustrated and animated characters.
Test 1: a photo and a voice
Input: a portrait of a fictional woman generated with Kling 4.0 Image and a 14.9-second AI voice-over from ElevenLabs. I deliberately used Polish, a language full of clusters like “szcz” and “źdźbło” that are hard for lip sync - the sentence includes the tongue twisters “szczęście, źdźbło, chrząszcz”. Avatar 6 does not translate anything: it moves the lips to whatever audio you give it, in any language. Motion prompt: natural friendly talking, small head movements. 1080p.

Avatar 6 (regular) - 8/10
The lips follow the recording through the whole clip. On the hardest words they round clearly, and in the longer pause at about 6.9 s they close. In short breaks (0.9 s, 3.3 s, 11.7 s) she smiles with visible teeth instead of closing her mouth. File ready after 55 s, cost USD 1.48.

Avatar 6 Turbo - 7.5/10
Turbo moves the head more and articulates almost as precisely. The difference shows in the pause after 6.9 s: the lips stay slightly parted. Turbo was not faster for me - the file arrived after 83 s - but it cost half: USD 0.74.
Test 2: an illustrated mouse from a reference video
Input: a 5-second video made from a paper-craft mouse illustration (Kling 4.0 Image) and a 6.1-second AI voice of a cheerful storyteller.
The mouse talks, blinks and changes expression, and the paper-craft style stays intact - the model does not try to turn it into a realistic animal. I see no meaningful difference between regular and Turbo here, so for cartoon characters I would pick Turbo.
Avatar 6 vs OmniHuman 2 vs H3 Max Lip Sync
Same photo, the first 7 seconds of the same recording, default settings, all through fal.

| Model | Score | Resolution | Time to file | Price on fal |
|---|---|---|---|---|
| MiniMax H3 Max Lip Sync | 8.5/10 | 480×832 | 15 s | USD 0.05/s (480P) |
| HeyGen Avatar 6 | 8/10 | 1080×1920 | 55 s | USD 0.10/s |
| OmniHuman 2 | 8/10 | 720×1278 | 120 s | USD 0.105/s |
| HeyGen Avatar 6 Turbo | 7.5/10 | 1080×1920 | 83 s | USD 0.05/s |
H3 closed the mouth best in pauses and is the cheapest, but only in 480P. Avatar 6 and OmniHuman 2 tied: Avatar 6 gives sharp 1080p and livelier expressions, OmniHuman 2 a calmer, more natural face in 720p. The Avatar 6 times are for the full 14.9-second clip.
Price: fal or the HeyGen app
| Option | Price | Avatar 6 |
|---|---|---|
| fal, Avatar 6 | USD 6 per minute (1080p), USD 8.70 (2K) | all 3 modes |
| fal, Avatar 6 Turbo | USD 3 per minute (1080p), USD 5.25 (2K) | all 3 modes |
| HeyGen Free | USD 0 | 3 videos a month up to 1 min, watermark |
| HeyGen Creator | USD 29/month (USD 24 billed yearly) | Avatar 6 Turbo |
| HeyGen Pro | USD 49/month | Avatar 6 and Turbo |
| HeyGen Business | USD 149/month + USD 20 per extra seat | Avatar 6 and Turbo |
The price of one month of Creator (USD 29) buys 9 min 40 s of Turbo video on fal; one month of Pro (USD 49) buys 8 min 10 s of regular Avatar 6. If you make less than that, fal is cheaper. More on plans, modes and how to start: HeyGen - the complete guide to AI avatars.
How I tested
- 17 November 2026, fal.ai API, one run per mode, the same photo and voice for the comparison.
- Characters generated by AI (Kling 4.0 Image), voices from ElevenLabs. No real person.
- Scoring 0-10: do the lips open on syllables and close in pauses, artefacts on teeth and lips, natural expression. I checked the whole clip and mouth frames every 0.6 s against audio loudness.
- Clips on this page are compressed for preview; the originals from fal were 1080×1920.
FAQ
Does HeyGen Avatar 6 work in my language?
Yes. The model moves the lips to the recording you upload, so the language of the audio is the language of the video. In my test it kept sync even on Polish tongue twisters.
How much does a minute of Avatar 6 cost?
On fal in 1080p: USD 6 regular, USD 3 Turbo. In 2K: USD 8.70 and USD 5.25.
Can I animate a photo of a real person?
Only with that person’s consent and a clear label that the video is AI-generated. HeyGen requires consent confirmation for realistic avatars in its app, and the reference-video mode on fal does not accept videos of real people at all.
Sources
- My own test, 17 November 2026: 6 clips through the fal.ai API.
- fal.ai model pages and pricing: Avatar 6 Image to Video, Turbo Image to Video, Reference Video to Video, Identity to Video, OmniHuman v2, H3 Max Lip Sync (read 17 November 2026).
- HeyGen pricing and API documentation (read 17 November 2026).
The link to ElevenLabs is a partner link: if you buy through it, QUANTSLANT earns a commission and you pay the same price. It has no influence on the content, including where the coverage is critical.
More on this from QUANTSLANT
guides·MAI Image 2.7 review: Microsoft’s new #1 image editor, testedMAI Image 2.7 is #1 on Artificial Analysis for image editing (1215 Elo). I tested sign edits, photo restoration and style changes - plus price per image.