Lip sync AI API options: photo or video input, limits, price

A lip sync AI API takes a face and an audio track and returns a video whose mouth moves with the audio. The options split by what the face is: a still photo or avatar that gets animated (Creatify Aurora, HeyGen Audio to Video, and Sume's VEED Fabric 1.0 and MiniMax H3 Max Lip Sync), or an existing video whose mouth is redrawn to new audio (HeyGen Lipsync, Kling Lip Sync). Model hosts such as fal and Replicate list many lip sync models, each with its own price.

Vendor facts come from each vendor's own pages, read on 2026-09-28. Sume's come from the Models overview and the request schemas in the Sume API reference.

Which lip sync APIs take a photo, and which take a video?

What about fal and Replicate?

Both are model hosts that list lip sync models from several model makers. fal lists them under a lipsync category and, per its pricing docs, bills each model by its own unit, only for successful outputs. Replicate's lipsync collection covers models that sync lips in videos or images to new audio; its pricing page says some models bill by hardware time and others by input and output. Read the model's own page for its inputs and price.

How do I choose a lip sync API?

• Start from what you have. A still photo or an avatar needs a photo-driven API; a finished video you want to redub needs HeyGen Lipsync, Kling, or a hosted video-to-video model. Sume's two endpoints take a still or an avatar, not existing footage.
• Check where the audio must live. Sume refuses audio that is not on its media host, so the usual source is a Sume TTS clip; there is no public upload for a local file. HeyGen takes a public URL or an upload, and Kling a URL or Base64.
• Check the length caps against your longest line: 2–60 seconds on Kling, 5–14.8 on H3 Max, up to 300 on Fabric, 5 minutes on Aurora, 30 minutes on HeyGen Audio to Video.
• Compare the unit: per second, per 5 seconds, per minute or per credit. Sume's rates are before a 5.5% agent fee by default.

Can a video model lip-sync to my voice-over instead?

Not on Sume. Its models guide says video models do not lip-sync to generated TTS or to a later voice-over, so a talking face is built from a still plus audio on Fabric. The request, limits and a worked price are in Lip sync API: image and audio to a talking clip, and lip sync vs dubbing explains when you need which.