ElevenLabs Lip Sync at 480p or 720p vs Sume's lip-sync routes

ElevenLabs' Lip Sync tool turns a chest-up portrait and a voice track into a talking video at 480p or 720p, in its web app and behind a paid plan. Sume sells the same idea as API routes billed per second: VEED Fabric at $0.1875 a second at 720p, and MiniMax H3 Max lip-sync at $0.0625, $0.10 or $0.20 a second at 480p, 768p and 1080p.

What ElevenLabs says

The ElevenLabs page describes three steps: upload a portrait with chest-up framing facing the camera, add speech by uploading audio or typing a script with an AI voice, and download the result. Output is MP4 at 480p or 720p, upscalable by up to 4x in the platform. It lists over 30 languages through voice synthesis and voice cloning.

Video generation needs a paid plan, credits are shown before generation, and the page gives no per-second price. A promotional 3x credit offer on Creator+ runs through October 12. There is nothing on the page that lets us convert credits into dollars, so this post does not.

What Sume lists

Both Sume routes take one still image plus audio, create a job of type avatar_image_to_video and use the standard job lifecycle. Fabric is the default talking model. H3 Max lip-sync is the explicit alternative when the audio is between 5 and 14.8 seconds. Audio must be hosted on Sume media and is limited to 10 MB.

H3 Max price is ceil(duration) x the provider list per second x 1.25: $0.05, $0.08 and $0.16 list become $0.0625, $0.10 and $0.20 at 480p, 768p and 1080p. Fabric at 720p is $0.1875 a second.

A 10-second example

A 10-second clip on Fabric at 720p is 10 x $0.1875 = $1.875. On H3 Max at 768p it is 10 x $0.10 = $1.00, and at 1080p it is 10 x $0.20 = $2.00. If the audio is under 5 seconds, H3 Max refuses it, and Fabric handles it.

Which to pick

• You are one person making a few clips: a web app with a plan is fine.
• You need a clip per row of a spreadsheet, with a job id and an idempotent retry: use the API.
• Your audio is already a file: the lip-sync routes. If you only have a script, an Avatar 1.0 talking video speaks it for you.

What to do

Run the same 10-second voice file through both and compare mouth shape on a side profile before you commit. Check the live rate in the Sume catalog before a batch.