Metahuman Image Rights with AI Models

Development Cinematics & Media
,
1

I’ve a quick question about linear media export of Metahumans in regard to using those images in AI-generated imaging tech.

My workflow currently includes sculpting my own character and using the Metahuman plugin that converts the mesh to Metahuman for texturing and rigging. I then take that character back to UE5 and render them out in single images in different poses, clothing, lighting, and environments. I then take those images of the character and train my own version of Stable Diffusion so that I can prompt that character in different scenarios.

I plan to use my images for book illustrations.

Originally I started considering this workflow because the EULA said this;

b. Linear Media (e.g., Rendered Video Files)
You may freely Distribute Licensed Content incorporated into rendered linear media products. This means, for example, you may freely Distribute rendered video files (e.g., broadcast or streamed video files, cartoons, or movies) or images created using Licensed Content.

Now that I’m ready to build a usable model for my project I went back and checked the Eula to make sure everything was kosher and saw these addendums that confused me.

(You May Not) d. use MetaHuman Content for the training or testing of any artificial intelligence, machine learning, deep learning, neural network or similar technology.

MetaHuman Content includes both digital human characters created by your use of MetaHuman Creator and digital human characters created by or on behalf of Epic.`

I’m not planning to use the actual 3D models or Metahuman rig in my AI models. I’m using linear renders to train and accomplish my goal. Am I allowed to do this? Or would the rendered media fall into the same definition as “Metahuman Content” and prevent me from doing this? I guess the question is what is the definition of a Metahuman Character? Is it the likeness or the 3D model?

It’s totally fine if I can’t do this - I can build my own characters. Metahuman Creator in the workflow just sped up the process by a few days.

Thanks for any thoughts/insights.

2

I asked about this originally. I was told unofficially, by a back channel, that this means SYNTHETIC DATA generation. I think the exact method you are describing is not the original intention of this caveat. From a purely copyright point of view, in regard to still images, if you have made any custom alteration at all (which the system is intended for), then I don’t see how any copyright could stick in this case.

Keep in mind, they can’t get blood from a stone. If you are not making money hand over fist, they are not likely to pay a lawyer to try and get some of it.

For more discussion of this kind of thing refer to the Virtual Beings Facebook group.

3

Thank you for responding. I will do some more research on Synthetic Data Generation and take a look at the FB group. I appreciate your time.

4

Hi, sorry i am very slow on legal things. I am trying to render some image sequences with different camera angles on a animated metahuman and use it for training 3d gaussian splatting pipelines or NeRF. Th pipeline will eventually be transferred to produce real human result, and I do want to publish an academic paper on this. Does their EULA prohibits me from doing so? Quoting from UE:

"MetaHuman and the MetaHuman assets may not be used for the purpose of building or enhancing any database; training or testing any artificial intelligence; machine learning; deep learning; neural networks; or similar technologies. "

but I also saw a metahuman dataset being published on kaggle, which confused me further.