Shipping a Real-Photo Editing Feature Responsibly

Your face-generation product is adding a feature: users upload their own photo, the system inverts it into the generator's latent space, and the user can then apply attribute-direction edits (age, expression, lighting) and download the result. Leadership wants this shipped in four weeks and asks you to sign off on the system design.

1. What does the generation service need to store beyond the final output image, and why does this matter specifically for this GAN- based feature (as opposed to, say, a diffusion-based image editor)?
2. Name two concrete, non-negotiable product/policy decisions this feature raises that a purely technical design review would miss if it only reviewed the inversion and editing pipeline.
3. A stakeholder says "we don't need a moderation stage — GAN inversion just reconstructs the user's own uploaded photo, edited slightly, so there's nothing new to moderate." Evaluate this claim.