I am mostly interested in using my own lyrics - for two reasons: I do write them and the lyrics I have seen from the OTHER AI song makers are tripe. First attempt confused the hell out of the machine by not giving enough text I think, just went weird not singing what was showing and then when tried to extend, ignoring the new text.

However second attempt was a full lyric, three verses. The result was not that close to the parameters of the prompt in terms of style but was really really good. I then grabbed the result - not by "download" which did not work but by making a video and stripping the sound. Then I orchestrated this in Logic. I love the result. This kind of inter-operability with existing musical practice will not make you more money, but it is a way forward that is non-trivial, meaningful to musicians not just the wishful amateur. I am not being contemptuous of non-musicians doing whatever with such tools - it is revolutionary and meaningful even if the results are variable.

I have more ideas about how this path might work if you are interested in that direction. Or you can just do the same as Udio etc but better (you are already better) and then monetise that. Why not both ?

Questions: Is my now altered, orchestrated version my song? Can it be released? How do I get to use the same "singer" (voice model) for another song? Is there a way you could accept audio (song parts) input as well as text prompt?

UPDATE

Next attempt using a lengthy, rather literary poem/lyric produced a fair bit of spoken word, seeking to fit the word into the 1'30 first pass limit. Interesting but not what I wanted - though the English accent male version I could use. Then broke it in half - first 18 lines as two "verses" - this was more successful - "orchestral song" taking the result very much in music theatre territory. Unfortunately extending and trying to add the second half of words (another 17 lines) the response was to rush through them as spoken word!! So I just made another "new" song with the second half words. Bothe versions of this were male voice (whereas first was one man one woman) and really good. Second of two options included some gobbledegook but I quite liked it. It got back onboard track then weird again but the climax with big male voice harmonies was impressive and truly weird. This could be the beginning of the first AI Broadway show. The first attempt is also good with no glitches:

https://sonauto.ai/song/2df1b36b-5af6-45fd-939c-779fd67e829c

However my attempts to extend just the music repeatedly added more words - not mine - not so great, so this just ends abruptly for now.

I am generally impressed and boggled.

However to give you useful feedback the handling of lyrics is awkward - extending and editing them is not intuitive and unpredictable in effect. The response to original lyrics is variable - some hallucination and weirdness, some genuinely beautiful and creative melodising. Amazing really. Also I could seem to have a prompt AND specify styles and put in custom lyrics - they are separate pages and did not always "stick". Consequently I chose the custom lyrics over experimenting with prompts which I REALLY need to do.

Cheers. I will eventually post these results somewhere.

Chris