Skip to main content

​
Choosing a model

SongCraft generates full songs from a prompt. MiniMax has notable audio depth too, and Mureka is worth comparing.
ElevenLabs is the default for text-to-speech and voice work.
Filter for audio-to-audio. ElevenLabs and MiniMax both offer it.
Look at text-to-audio models — they cover effects as well as music. Read the model descriptions, since capability names don’t distinguish the two.
Audio quality is subjective in a way image quality isn’t. Generate the same prompt on two or three models and listen. There’s no substitute for that.

​
Voice and video together

Two steps. Generate the voice track (Generate Voice chained action or ElevenLabs), then sync it — Add Lipsync on an image, Lipsync Video on a clip, or Sync directly.
Yes — audio-to-video models generate video from an audio track. Kling and LTX both have them.
HeyGen handles presenter video directly, rather than assembling voice and lipsync yourself.

​
Costs

Often per second, like video — so a long track costs proportionally more. The cost is shown before you submit. See Credits & Plans.
Generate a short clip to check the voice or musical style before committing to full length. Style is audible in a few seconds.

​
Practical

My Assets → Generated, alongside images and video. Download from there or from the preview.
Yes — upload it in the form or pick it from your assets, for lipsync, audio-to-video, or audio transformation.
Yes — the Video Editor assembles clips and audio. Generate the pieces, then arrange them there.
Check status — audio is its own service group. Failed runs aren’t charged.