Kling AI Avatar v2 Pro on fal: A Talking Photo for $0.115 per Second (October 2026)
One photo and one audio file become a talking video as long as the audio, at $0.115 per second on Pro and $0.0562 on Standard. What it accepts, how it behaves with real faces, and when to choose it over OmniHuman 2.0 or HappyHorse.
Kling AI Avatar v2 is Kling's talking-avatar model: you give it a portrait and a recording, and it returns a video of that person delivering the recording. It comes in two tiers on fal, Pro at $0.115 per second and Standard at $0.0562, and it is the talking-photo model that has been in front of the most real users in our own product, because we used it to build our onboarding video in early September. This guide covers what it takes, what it costs, and where it fits among the newer options.
What it costs
| Audio length | Pro ($0.115) | Standard ($0.0562) |
|---|---|---|
| 10 seconds | $1.15 | $0.56 |
| 30 seconds | $3.45 | $1.69 |
| 1 minute | $6.90 | $3.37 |
| 5 minutes | $34.50 | $16.86 |
Prices read from fal on 18 October 2026. The output is as long as the audio, so cost is simply audio length times the rate.
What it takes
- An image of the speaker (
image_url). The video inherits the photo's aspect ratio, so supply a 9:16 portrait if you want a vertical result; the model does not reframe. - An audio file (
audio_url). Length can run to about five minutes. - A prompt describing the delivery. This is where most people go wrong, as the next section explains.
It does not generate speech. If you have a script and not a recording, you need a voice first, either your own recording or a text-to-speech output you have rights to.
The two things that decide quality
A restrained prompt. "Speaks calmly to camera, slight head movement" holds a face. "Gestures energetically while laughing" does not. This is the highest-leverage change available and it is free.
A clean photo. Even frontal light, the whole head in frame, no heavy shadow across one side, no sunglasses. The model infers a three-dimensional head from the still, and ambiguity in the input becomes wobble in the output.
Keep individual clips short, five to eight seconds where you can, and cut several together for anything longer. A cut is invisible; drift is not.
Pro or Standard
Our rule: Standard for anything watched small or produced in volume, Pro when the face fills the frame. Standard is about half the price and a real talking-avatar model, not a demo tier; Pro buys visibly better facial fidelity.
Where it sits among the alternatives
| Model | Price per second | Best at |
|---|---|---|
| Kling AI Avatar v2 Standard | $0.0562 | volume, small-screen viewing |
| MiniMax H3 Max Lip Sync | $0.05 (768P base) | very long audio, up to 15 minutes |
| OmniHuman 2.0 | $0.105 | body and gesture that follow the audio |
| Kling AI Avatar v2 Pro | $0.115 | a face that fills the frame |
| HappyHorse 1.1 image-to-video | $0.14 (720p) | generating the voice from a script in one call |
The decision runs on what you hold. A script and no recording: HappyHorse writes the audio itself. A recording and a modest budget: Kling Avatar. A recording where the delivery carries the joke or the persuasion: OmniHuman. Our earlier comparison of HappyHorse and OmniHuman explains the script-versus-audio split in depth.
What it does not do
It animates a still; it does not put an animated head into existing footage. For that, and for fixing a mouth in a video you already shot, see lip-sync and video-edit models (sync-3 and HeyGen compared). And it is not a source of voices: the responsibility for the recording, including its rights, is yours. Cloning a real person's voice and putting words in their mouth is the version of this format that gets reported.
Our experience
We chose Kling Avatar v2 Pro for the onboarding video because it accepts a real face without the policy refusals some other video models give, because the cost of a short clip is small, and because the output holds when the prompt is restrained. We have not benchmarked it against OmniHuman 2.0 or H3 Max Lip Sync on identical material, so the table above is a price and capability comparison, not a quality ranking.
Common questions
Does it generate speech?
No. It animates a portrait to match a recording you supply. For a script without audio, you need a voice first or a model that writes its own audio, such as HappyHorse image-to-video.
How long can the audio be?
About five minutes on fal. Output length equals audio length.
Pro or Standard?
Standard for volume and for anything watched small, Pro when the face fills the frame. Standard is about half the price and a real avatar model, not a demo.
How do I keep the face stable?
Start from an evenly lit frontal photo, use a restrained prompt such as "speaks calmly, slight head movement", and keep clips short, joining several for anything long.
How to do this without code: A talking portrait from a recording is the Animate photo tool with audio. In AutoHustle this is the Animate photo tool with a voice recording: audio in, a talking photo out (about 3 credits per second with the audio-driven model, quoted before you press Generate). See How to make a talking photo and Animate a photo.
Related reading: Kling LipSync Audio-to-Video on fal: Limits, Price Unknowns and Why Reels Need a Scene-by-Scene Approach (October 2026) · FlashTalk on fal: A $0.02-per-Second Talking Head That Cannot Make a Vertical Video (October 2026)
For developers: the fal endpoints
| Model | fal endpoint |
|---|---|
| Kling AI Avatar v2 Pro | fal-ai/kling-video/ai-avatar/v2/pro |
| Kling AI Avatar v2 Standard | fal-ai/kling-video/ai-avatar/v2/standard |
Inputs: image_url, audio_url, prompt. Output length equals the audio length; aspect ratio comes from the image. Prices read on 18 October 2026.