Leaderboard

I mean, OP doesn't mention "basic prompts" or anything of the sort, it just mentions prompts. The "simple" part was by me 😋 There's really 2 points in the OP, one is about the ethics of how the models were trained regarding artist consent and copyright. I'm not sure if OCR's stance would be the same if the models were "ethically" trained or if, for example, a user trained their own model only on songs they made or something like that. There's a second point about the "interpretation" aspect of AI generated music, the "human touch". I think we simply disagree here, I don't think current tools provide much "human touch". Even if Suno allows for more than "a simple prompt", the "human part" is still a low percentage of the finished creation. Of course this could change with time and how the technology evolves but it's my stance on what I've seen of current technology. In the end, it's up to the community to decide how it approaches these things. I think it's fair if OCR decides it wants to focus on music that is created on a larger percentage by the "human touch" and, of course, policies can change over time as technology evolves.

I've been thinking about this. I had a look at the submission standards. I think they need to be rewritten for future submissions. 2.1. Your submission must be your own, original arrangement. 2.4. If your submission involves multiple artists, ensure that all artists: (are named, agree with the content policy, and are credited how they want to appear on ocr) Without any opting out of being part of the training data of some of the ai generators, and some of them (Udio at least) being sued over recreating copyrighted tracks, it's difficult to confirm that an arrangement is the artist's own, original arrangement. And while sample libraries and loops are licensed and don't need crediting, the artists contributing (without consenting) to the ai generators are contributing their arrangements and performances to the tool, which is turn generates (possibly overfitting) an arrangement based on some amount of prompt and directing. Would using a ghost writer for the arrangement and not crediting them (anywhere) be acceptable on ocr? I think that's a close enough analogy to some of what the ai generators do. 3.3. Any incorporation or arrangement of source material not from games (mainstream, classical, etc.) should be extremely limited. This isn't just something an ai generator might do, but when it does so, it might do so deliberately, as in referencing the track in the training data with the intent of recreating some of it. When a human does it, it can be coincidental, simply doing the logical thing based on a chord progression or rhythm, not having ever heard the thing it resembles. If I recognize that my remix is ripping off the 5/4 bass and brass combo from Lingus, I can rewrite it. But what if an ai generator recreates it and the artist doesn't recognize it because the artist doesn't know it? Can it be someone's own, original arrangement and be built on and contain sizable portions of someone else's track without the artist knowing it? 4.2. Your arrangement must be substantial and original. Submissions must be different enough from the source material to clearly illustrate the contributions, modifications, and enhancements _you_ have made. My emphasis. The ghost writer scenario applies here too. Different from the source can't just mean a close copy of some other track that exists in the training data instead. 5.2. Production must show significant attention to sound quality, mixing, mastering, and utilization of effects. Whose attention? Logic's gotten its new ai mastering tool (haven't upgraded and tried it yet). Isn't Neutron some kind of auto-mixing tool? Tools are tools, and it's difficult to draw a line exactly in between when a tool is convenient and when it does everything for you. I don't think ocr would reject a track for using Neutron or something. But at some point a tool is less a tool and more a collaborator. At that point, credit is due. And given how the generators are trained, and the black box between training data and output, how do you do that? I suppose a new section regarding ai generators should be added regarding where ocr draws the line. Don't know what it should contain. Do all notes need to be inputted by the artist(s)? Does the artist need to have had full control over each track in the mix? How does using an ai performance of the artists' notes count? What about non-neural-network automation tools that humanize things? -- My general stance is that it should be clear what an artist has done. Like in point 4.2. I know I get disappointed when I find out that the cool thing I like about a track is just a loop anyone could have used. What was impressive about the track and the artist is then just an asset they found. But that doesn't mean they can't use the tools available to them. A guitar player is using a tool to play the notes, more notes than they could produce with just their mouth. Usually. A composer isn't necessarily performing, and a performer isn't necessarily composing. People often select their sounds from presets rather than make their own. There are automated mastering tools, be they based on written algorithms or neural networks. There's no shortage of tools that simplify the music-making, but we all have some idea of what counts as effort and skill, and even if my performance on Eye of the Storm isn't all that interesting (if any performance at all), my choice of notes and sounds is what makes that track mine. I think that's the key. What did the artist actually do? What's the thing that makes this _theirs_? A full ban on everything related to ai is probably not the right direction, when much of the same criticisms of making things too easy can be said for automation like humanizing features in sample libraries, chord generators, arpeggiators and multiband compressors. And synths. And Hatsune Miku. And at some point we might turn into goat farmers. But there's the other direction too, where the artist is less creator and more facilitator. Director. Producer. Not performer, writer, designer. Patron rather than artist. I don't know if there's a convenient stopwatch-like criterion that can be used, so I think the best I can come up with, for now, is the question regarding an artist's work: "What makes this track theirs, and not the tool's?"

Hey everyone, UPDATE: My version of Dan's theme. Same rules as above - made almost entirely using the SFA soundfont. Enjoy!