Post #3974525

2026-07-21 00:25 UTC

@deFractal@infosec.exchange I am thinking that such legal language could be really difficult. My sense is that we would want to allow certain image manipulations - such as color balance or dust removal. (My own skills at color balance are below abysmal, so I look at tools that do that as wonderfully assistive forms of magic.) Could cropping, creative lighting, or even compositional posing of a photo be considered improperly manipulative? Also, I saw someone mention that because many animals see a different range of color/frequencies than we humans do that image processing that constrains or changes non-human visible colors might be either "interesting", "objectionable", or "a new way to do steganography.

Replies (1)

@deFractal@infosec.exchange 2026-07-21 01:37

@karlauerbach@sfba.social Yes, it would take careful legal–technical definition of “generative AI” (or, to generalize [Dr. Bender](@emilymbender@dair-community.social)'s more apt term for LLM output, synthetic media extrusion). If a transformation uses matrix or tensor values that are some combination of geometrically or formulaically computed constants and the user-provided input media, then to my first approximation, it could include such things as colour balance or dust removal filter or the like, and would not constitute synthetic media. If it uses matrix or tensor values derived from wholesale scraping and industrial-scale gradient descent (or functionally similar computations), it's a plagiaristic media synthesis model, and would be reason not to add a legal declaration of absence of synthetic media use, as suggested in my OP, in order to avoid a variety of grounds from civil suit and (for example) perjury, fraud, DRM circumvention, or computer/network intrusion charges (depending on the creativity of the legal word smithing and perhaps accompanying perfunctory DRM or IAM mechanism tied to cryptographic keys derived from said declaration).

Open ##3974524