ugc-diversity-representation
Installation
SKILL.md
Representation in Generated Creative
Image and video models have strong demographic defaults. Ask for "a person using a product" repeatedly and you get a narrow band back — typically young, light-skinned, slim, non-disabled, and conventionally attractive. That's a property of training data, not a choice you made, but it becomes your choice the moment you publish it.
Generation removes the usual excuse. Casting a varied roster costs a few credits.
The default is narrow, so specify
The practical mechanic: unspecified means default. If you don't name age, body, skin tone or ability, the model picks, and its pick is not representative.
Weak: "a woman using the product in a kitchen"
→ the same narrow output, every time
Specified: "a woman in her fifties, mid-size build, deep brown
skin, short grey-flecked hair, ordinary appearance,
plain jumper, no makeup"