# Talking Head API Pricing 2026: 5 Photo-to-Video APIs Tested, 35 Prices Checked

> Stork sent the same photo and 20-second voice clip to five talking-head APIs (Kling AI Avatar v2, OmniHuman 1.5, VEED Fabric, Stork Standard and Pro), scored lip sync with SyncNet and compared motion, speed and price, plus 35 per-minute API prices read on providers' own pages. November 13, 2026.

- Source: https://www.stork.ai/blog/talking-head-api-pricing-2026
- Publisher: Stork.AI editorial team
- Published: 2026-11-14
- Updated: 2026-11-14
- Language: en
- Reading time: 8 min

## Key takeaways

- All five APIs we tested had good lip sync (SyncNet 6.3 to 8.1); price did not predict it.
- What price buys is motion: head-only at $0.20 to $0.40/min, gestures and body motion at $9 to $9.60/min.
- 720p per minute: Stork $0.20, HeyGen Avatar III $0.99, Hedra $3.00, Kling $3.37, VEED Fabric $9.00, OmniHuman 1.5 $9.60.
- Alibaba's $0.17/min LivePortrait API was retired on October 10, 2026; older lists still quote it.
- Length caps vary from under 20 s (Alibaba wan2.2-s2v) and 60 s (OmniHuman 1.5 on fal) to no limit (Stork).
- The cheapest model was also the fastest in our run; Kling took almost 7 minutes for 20 seconds.

**Short answer:** we sent the same photo and the same 20-second voice clip to five photo-to-talking-video APIs and scored the results. Lip sync was not what separated them: every model landed in the "good" band of the SyncNet lip-sync judge (6.3 to 8.1). What the price buys is **motion**. The two cheapest models move only the head and face, Kling adds small hand movements, and OmniHuman 1.5 and VEED Fabric move the body and gesture with the hands. On price, the spread is roughly 50x: from $0.20 per minute of 720p video (Stork Standard) and $0.99 (HeyGen Avatar III) to $3.00 (Hedra Character-3), about $3.37 (Kling AI Avatar v2), $9.00 (VEED Fabric) and $9.60 (OmniHuman 1.5 on fal).

**What this is:** Stork's own test, run on November 13, 2026, plus a price check of 35 API offers read on each provider's own pricing page the same day. **Disclosure:** Stork sells one of the APIs in the test ([Stork Talking Photo API](https://stork.ai/developers/talking-photo-api), built on the open [SoulX-FlashHead](https://github.com/Soul-AILab/SoulX-FlashHead) model). We report its numbers the same way as everyone else's, including where it loses.

## Test results

| API (model) | Price per minute, 720p | Lip sync (SyncNet, higher = better) | Time to a 20 s video | What moves | Output |
| --- | --- | --- | --- | --- | --- |
| Stork Standard (FlashHead Lite) | $0.20 | 6.85 | 96 s | Head, face, eyes | 982x720 |
| Stork Pro (FlashHead Pro) | $0.40 | 6.55 | 143 s | Head, face, eyes | 982x720 |
| Kling AI Avatar v2 Standard (fal) | $3.37 | 6.35 | 412 s | Head, face, small hand movements | 1120x816, 30 fps |
| OmniHuman 1.5 (fal, 720p) | $9.60 | 6.87 | 336 s | Head, shoulders, body, hands | 1120x832 |
| VEED Fabric 1.0 (fal, 720p) | $9.00 | 8.11 | 135 s | Head, body, expressive hand gestures | 1120x832 |

*Same photo and 20.06 s of audio for every API, one run each, November 13, 2026. Time = submit to file ready, including queueing.*

**How to read the lip-sync column:** SyncNet's confidence score (LSE-C) measures how well mouth movement matches the audio. In our calibration on singing and speech, 5.4 to 7.8 is good and under 1.5 is broken. All five are good; VEED Fabric scored highest. All five also kept the audio offset within 2 frames.

![Stork Standard ($0.20/min): four moments from the same 20 s clip. Head and face move; hands stay still.](https://expert-egret-407.convex.cloud/api/storage/ac2338b2-d61b-4ce2-b557-feed062206af)

![Stork Pro ($0.40/min): steadier face, same framing. Body and hands stay still.](https://expert-egret-407.convex.cloud/api/storage/e170e0cd-7c89-4de8-bd4f-ddc297b07472)

![Kling AI Avatar v2 Standard ($3.37/min): small hand and shoulder movements.](https://expert-egret-407.convex.cloud/api/storage/f08866d6-01fe-4f4b-ba21-82a25b8b9c9a)

![OmniHuman 1.5 ($9.60/min): the body shifts, leans and breathes; hands move on the desk.](https://expert-egret-407.convex.cloud/api/storage/09c0731f-539d-41bc-a5f4-972449e75725)

![VEED Fabric 1.0 ($9.00/min): the most animated, with open-hand gestures.](https://expert-egret-407.convex.cloud/api/storage/dd92edd4-ad20-4fec-9acc-27c8ed272161)

## What you are paying for

- **Lip sync is solved at every price.** None of the five drifted out of sync over 20 seconds, and the cheapest scored within 0.1 of OmniHuman.
- **Body motion is the expensive part.** A head-only model like Stork's animates the face region and keeps the shoulders and hands frozen. Diffusion models such as OmniHuman, Fabric and Kling regenerate the whole frame, so the person can gesture, at 15 to 50 times the price.
- **Speed does not follow price.** The cheapest model was the fastest, and Kling, mid-priced, was the slowest at almost 7 minutes for 20 seconds of video.
- **Length limits differ a lot.** OmniHuman 1.5 on fal is capped at 60 s per call at 720p, Alibaba's wan2.2-s2v at under 20 s, Kling at 300 s, Hedra and WaveSpeed's InfiniteTalk at 10 minutes, WaveSpeed's FlashHead at 30 minutes. Stork has no length limit.
- **The same open model is sold at very different prices.** SoulX-FlashHead, the model behind Stork, costs $0.30 per minute on [fal](https://fal.ai/models/fal-ai/flashhead) (text in, 512x512 head out, its own voices only), $0.90 to $1.80 per minute on [WaveSpeed](https://wavespeed.ai/models/wavespeed-ai/soulx-flashhead) (your photo and audio) and $0.20 per minute at 720p on Stork (your photo and audio, pasted back into the full photo).

## Which one to use

- **Long narration, faceless YouTube channels, course lessons, news recaps, dubbed versions of one presenter:** a head-only model is enough and the price difference compounds. A 10-minute video costs about $2 on Stork Standard against about $33.70 on Kling and $96 on OmniHuman 1.5.
- **Short ads, hooks, UGC-style clips where the presenter should feel alive:** VEED Fabric or OmniHuman. The hand gestures are worth paying for in a 15-second ad.
- **A middle ground:** Kling AI Avatar v2 moves a little more than a head-only model at about a third of OmniHuman's price.
- **Real-time conversation (agents, kiosks):** none of these. Rendered APIs take minutes; live avatars are a separate product category priced per session minute.

## Every API price we checked, cheapest first

Every row below was read on the provider's own pricing page or docs on November 13, 2026, and converted to dollars per minute of finished video. "Estimate" means the per-minute figure is our math from a plan's published credits.

| Provider and model | Resolution | Per minute | Notes |
| --- | --- | --- | --- |
| Stork Standard ([API](https://stork.ai/developers/talking-photo-api)) | 480p / 720p / 1080p | $0.10 / $0.20 / $0.25 | Any length, per started 10 s, no watermark, prepaid credits |
| Stork Pro ([API](https://stork.ai/developers/talking-photo-api)) | 480p / 720p / 1080p | $0.30 / $0.40 / $0.60 | Larger FlashHead model |
| [fal FlashHead](https://fal.ai/models/fal-ai/flashhead) (same open model as Stork) | 512x512 head | $0.30 | Text input only, read by built-in ElevenLabs voices; no own audio |
| [A2E](https://a2e.ai/pricing/) photo to avatar to lip sync | up to 2K | $0.30 to $0.59 (estimate) | Needs an avatar built first; Pro plan |
| [D-ID API](https://www.d-id.com/pricing/api/) (annual plans) | not stated | $0.69 to $0.78 (estimate) | Subscription; Launch plan carries an AI watermark |
| [WaveSpeed InfiniteTalk-Fast](https://wavespeed.ai/models/wavespeed-ai/infinitetalk-fast) | not stated | $0.90 | $0.015/s, up to 10 min |
| [WaveSpeed SoulX-FlashHead](https://wavespeed.ai/models/wavespeed-ai/soulx-flashhead) | 480p / 720p | $0.90 / $1.80 | Photo + audio, up to 30 min, 3 s minimum |
| [HeyGen API](https://help.heygen.com/en/articles/10060327-heygen-api-pricing-explained) Avatar III photo avatar | 720p/1080p | $0.99 | Plus $1.32 to create each photo avatar |
| [Replicate wan-2.2-s2v](https://replicate.com/wan-video/wan-2.2-s2v) | not stated | $1.20 | $0.02/s |
| [Magic Hour](https://docs.magichour.ai/billing/overview) talking photo (prompted) | 480p/720p | $1.20 to $1.90 (estimate) | Plan needed, 60 s max |
| [Replicate p-video-avatar](https://replicate.com/prunaai/p-video-avatar) | 720p | $1.50 | $0.025/s |
| [Hedra Character-3](https://www.hedra.com/models/video/hedra/character-3) | 540p / 720p / 1080p | $1.50 / $3.00 / $3.75 | Up to 10 min |
| [A2E](https://a2e.ai/pricing/) Talking Photo V1 | up to 2K | $1.80 to $3.56 (estimate) | 6 credits per second |
| [HeyGen API](https://help.heygen.com/en/articles/10060327-heygen-api-pricing-explained) Avatar IV photo avatar | 720p/1080p | $2.31 | No subscription |
| [Kling AI Avatar](https://kling.ai/document-api/pricing/base/video) standard (direct, fal, Replicate, WaveSpeed) | 720p | $3.36 to $3.37 | Direct API needs a prepaid package |
| [Akool](https://akool.com/api-pricing) talking photo | 720p | $3.48 to $4.08 (estimate) | Annual plan |
| [BytePlus OmniHuman 2.0](https://www.byteplus.com/en/product/OmniHuman) | 720p | $4.20 | $0.07/s, 30 to 300 s |
| [Alibaba wan2.2-s2v](https://www.alibabacloud.com/help/en/model-studio/wan-s2v-api) | 480p / 720p | $4.30 / $7.74 | Beijing region only, audio under 20 s |
| [fal HeyGen Avatar 4](https://fal.ai/models/fal-ai/heygen/avatar4/image-to-video) | 720p | $6.00 | $0.10/s |
| [fal OmniHuman v2](https://fal.ai/models/fal-ai/bytedance/omnihuman/v2) | not stated | $6.30 | $0.105/s |
| [Kling AI Avatar v2 Pro (fal)](https://fal.ai/models/fal-ai/kling-video/ai-avatar/v2/pro) | 1080p | $6.90 | $0.115/s |
| [BytePlus OmniHuman 1.5](https://www.byteplus.com/en/product/OmniHuman) | not stated | $7.20 | $0.12/s |
| [Runware OmniHuman 1.5](https://runware.ai/docs/models/bytedance-omnihuman-1-5.md) | not stated | $7.92 | $0.132/s |
| [fal Creatify Aurora](https://fal.ai/models/fal-ai/creatify/aurora) | 480p / 720p | $4.20 / $8.40 | Rounded up to whole seconds |
| [fal VEED Fabric 1.0](https://fal.ai/models/veed/fabric-1.0) | 480p / 720p | $4.80 / $9.00 | Fast variant $6.00 / $12.00 |
| [fal OmniHuman 1.5](https://fal.ai/models/fal-ai/bytedance/omnihuman/v1.5) | 720p / 1080p | $9.60 | 60 s max at 720p, 30 s at 1080p |
| [fal LongCat Avatar](https://fal.ai/models/fal-ai/longcat-single-avatar/image-audio-to-video) | 480p / 720p | $9.00 / $18.00 | About 51 s max |

*Read on each provider's own page, November 13, 2026. Some providers list several models; the table shows the photo + audio route.*

> [!FACT] Alibaba's EMO and LivePortrait APIs, which were the cheapest photo-to-talking-video APIs anywhere (about $0.17 per minute for LivePortrait), were [retired on October 10, 2026](https://www.alibabacloud.com/en/notice/model_studio_notice_of_retirement_for_selected_legacy_longtail_models_7d4). Price lists and "best API" articles written before then still quote them. fal's InfiniteTalk, FlashTalk and MultiTalk endpoints are also marked no longer supported.

### Prices we could not verify

- **VisionStory's API rate.** Its [app pricing](https://www.visionstory.ai/pricing) works out to roughly $0.65 to $0.78 per minute at 720p, but the API is priced separately and not published.
- **Vidnoz's API rate** for talking photos (its pages disagree on credits per minute), **HeyGen's newest Avatar 6** photo-plus-audio price, **D-ID's** monthly prices and resolution, and the minimum prepaid package for **Kling's** direct API.
- **RunPod's InfiniteTalk endpoint** charges a flat $0.25 to $0.50 per video, but its maximum audio length isn't documented, so it can't be converted to a per-minute price.
- **Tavus, Synthesia, MiniMax and Pollo** publish no price for a single photo plus your own audio.

## Hidden costs to check before you pick

- **Setup fees per face.** HeyGen charges $1.32 to create each photo avatar before the per-minute rate applies; A2E's cheapest route also needs an avatar built first.
- **Minimums and subscriptions.** Kling's direct API is sold as prepaid resource packages, D-ID and Akool are sold as plans, and HeyGen stopped giving free API credits in February 2026. fal, Replicate, WaveSpeed and Stork bill per use.
- **Rounding.** Most per-second APIs round up to whole seconds; Stork rounds up to the next 10 seconds with a 5-cent minimum per video, D-ID to 15 seconds.
- **Length caps mean stitching.** A 10-minute video on a 60-second-capped API is ten calls plus your own stitching, and the face can drift between segments.

## How we tested

- **Inputs:** one photo of a presenter at a news desk (an AI-generated person, 1440x1056) and 20.06 seconds of English narration made with the open [Kokoro-82M](https://huggingface.co/hexgrad/Kokoro-82M) voice model.
- **APIs:** [Kling AI Avatar v2 Standard](https://fal.ai/models/fal-ai/kling-video/ai-avatar/v2/standard), [OmniHuman 1.5](https://fal.ai/models/fal-ai/bytedance/omnihuman/v1.5) at 720p and [VEED Fabric 1.0](https://fal.ai/models/veed/fabric-1.0) at 720p, all through fal's queue API with default settings, plus [Stork Standard and Pro](https://stork.ai/developers/talking-photo-api) at 720p. One run each. Hedra and HeyGen weren't tested because we had no API access.
- **Scoring:** lip sync with [SyncNet](https://github.com/joonson/syncnet_python) (confidence and audio offset), render time from submit to downloadable file, and four stills per video for the motion comparison above.
- **Cost of the test:** about $7.30 on fal, and a few cents of GPU for Stork's runs.

Try the cheapest option without code in the free [AI Talking Photo](https://stork.ai/tools/talking-photo) tool (one minute, with a small logo), or read the [API docs](https://stork.ai/developers/talking-photo-api). For apps that turn photos into talking videos for consumers, see our [best AI talking photo apps](https://stork.ai/blog/best-ai-talking-photo-apps-2026) guide.

## Frequently asked questions

### What is the cheapest talking head API?

Among prices published on providers' own pages on November 13, 2026, Stork's Talking Photo API was the cheapest API that takes your own photo and audio, at $0.10 per minute (480p) and $0.20 per minute (720p), with no length limit. fal sells the same open FlashHead model at $0.30 per minute, but only with text read by its built-in voices and a 512x512 output. Alibaba's LivePortrait API, about $0.17 per minute, was retired on October 10, 2026. Other verified pay-per-use APIs start at $0.90 per minute (WaveSpeed FlashHead at 480p and InfiniteTalk-Fast) and $0.99 (HeyGen Avatar III).

### How much does OmniHuman cost per minute?

OmniHuman 1.5 costs $9.60 per minute on fal and Replicate ($0.16 per second), $7.92 on Runware and $7.20 direct from BytePlus. OmniHuman 2.0 costs $4.20 per minute on BytePlus and $6.30 on fal. Prices read on November 13, 2026.

### How much does the Kling AI Avatar API cost?

Kling AI Avatar standard (720p) costs about $3.36 to $3.37 per minute of video ($0.056 per second) on Kling's own API, fal, Replicate and WaveSpeed. The Pro mode (1080p) costs $6.60 to $6.90 per minute. Kling's direct API is sold as prepaid packages.

### Do cheap talking head APIs have worse lip sync?

Not in our test. On the same photo and 20-second clip, all five APIs we tested scored in SyncNet's good range: Stork Standard 6.85, Stork Pro 6.55, Kling AI Avatar v2 6.35, OmniHuman 1.5 6.87 and VEED Fabric 8.11. The difference was body motion: the cheaper models move only the head and face, while OmniHuman and VEED Fabric move the body and hands.

### Which talking head API is best for long YouTube videos?

For long narration where a presenter mainly talks to camera, a head-only model is usually enough, and length limits matter more than motion. OmniHuman 1.5 on fal is capped at 60 seconds per call, Kling at 300 seconds and Hedra at 10 minutes. Stork's API has no length limit and costs about $2 for 10 minutes at 720p.

### Is there a free talking head API?

Not a free API in our check. HeyGen stopped giving free API credits in February 2026. Stork's Talking Photo tool lets signed-in users make two free one-minute videos a day with a small logo, and its API is pay-as-you-go from $10 of credits.