LTX-2.5 audio-to-video price vs Sume audio references

An llm-stats write-up reports that LTX-2.5 audio-to-video is billed on input audio duration, at $0.13 per second on Fast and $0.17 on Pro. Sume has no audio-to-video model id. Its reference_audio_urls field can accompany a clip, but it requires at least one reference image or video.

The reported prices

All numbers are from the llm-stats article.

Why billing on input seconds matters

When the price follows the input audio, the cost is set by your audio file, not by a duration setting on the video. A 30 second voiceover costs 30 seconds of audio-to-video, whatever else you ask. That makes the audio edit the cost lever.

What Sume offers

Sume's closest field is reference_audio_urls, which takes 1 to 3 audio URLs on the Video 1.0 alias and requires at least one reference image or video. On POST /v1/videos, the same idea is an input_references entry of an audio type. The Seedance 2.x models, Wan 3.0, MiniMax H3 and MiniMax H3 Max honor audio and video references. Gemini Omni Flash 1.1 does not take audio references.

That is a reference to guide a clip, not an audio-driven generator. Billing on Sume is by the model's own units at provider list times 1.25, so read pricing_skus for the model you choose.

If you need a voice-led clip

For a speaking face, Sume's route is a still plus audio through Fabric, which takes an audio_url and a measured duration_seconds. For narration over B-roll, generate the clips, then place the voice on a Timeline audio spine. Neither is an audio-to-video model, so they will not replace one if you want motion shaped by a soundtrack.