GochuJank
Member
- Messages
- 11
- Reactions
- Points
- 13
Hey all! I've been playing around a lot with blending morphs from existing characters, creating ones that are very different from the originals used. They do however have a LOT of dependencies, due to how they are sometimes even built on previous blends, sometimes with more than 50 pages of blendshapes. I have found a solution for personal use, which is to rebake the blendshapes so I can easily find them and remove the need for dependencies. But is there an official stance on this if I were to, for example, post free content here where some small part of the baked shape was originally derived from something I had paid for?
My personal interpretation of this comes down to likeness. Does the resulting character resemble the ones it is sourced from? If the likeness is too close then it is not considerably transformative and should not be allowed if from a source that disallows this. But if it is very different from the source material (ie. almost impossible to tell unless explicitly compared side by side), then I am not so sure.
I always believe in giving credit when creating derivative works, 100% if I can just have major dependencies like textures, clothing, hair etc. listed. But this workflow actually creates new data when it iterates. This means that as time passes and iterations occur, it becomes exponentially more difficult to give credit to every source - and one look could theoretically be derived from different weights of every single look ever made on the marketplace. Does this fall under fair use if the result is completely unrecognizable from its substituent parts? This usually boils down to whether the original authors are likely to lose revenue in the light of the new work, which some might argue is the case as the results are being published for free on the same marketplace. But this argument also hinges on how closely the new work resembles the original(s, plural in this case). It is SIGNIFICANTLY faster to make new, standalone looks this way vs manually sculpting so I am not going to write a to-do in here, as I do worry potential misconduct by less concerned parties.
However, I can't possibly be the only one who figured out how to do this as it is easier than you might think, so it is probably good to discuss sooner rather than later. From a practical standpoint, it seems the best solution is to not introduce paid content into my dataset. But from a legal standpoint it feels like a big mess of loopholes and subjective opinions, especially with the absence of AI authoring, as this is manual weight adjustment to create the new data - no AI involved.
My personal interpretation of this comes down to likeness. Does the resulting character resemble the ones it is sourced from? If the likeness is too close then it is not considerably transformative and should not be allowed if from a source that disallows this. But if it is very different from the source material (ie. almost impossible to tell unless explicitly compared side by side), then I am not so sure.
I always believe in giving credit when creating derivative works, 100% if I can just have major dependencies like textures, clothing, hair etc. listed. But this workflow actually creates new data when it iterates. This means that as time passes and iterations occur, it becomes exponentially more difficult to give credit to every source - and one look could theoretically be derived from different weights of every single look ever made on the marketplace. Does this fall under fair use if the result is completely unrecognizable from its substituent parts? This usually boils down to whether the original authors are likely to lose revenue in the light of the new work, which some might argue is the case as the results are being published for free on the same marketplace. But this argument also hinges on how closely the new work resembles the original(s, plural in this case). It is SIGNIFICANTLY faster to make new, standalone looks this way vs manually sculpting so I am not going to write a to-do in here, as I do worry potential misconduct by less concerned parties.
However, I can't possibly be the only one who figured out how to do this as it is easier than you might think, so it is probably good to discuss sooner rather than later. From a practical standpoint, it seems the best solution is to not introduce paid content into my dataset. But from a legal standpoint it feels like a big mess of loopholes and subjective opinions, especially with the absence of AI authoring, as this is manual weight adjustment to create the new data - no AI involved.
