face consistency is a solved problem in ai image generation — if you pay for it. it uses an identity-anchoring step (in lucy's case, pulid) layered on top of the base image model (flux dev).
most ai companion apps skip this step because it's expensive per-generation and because most users don't notice on the first photo. they notice on the fifth. by then the product has already locked in the “different woman every time” problem.
lucy's bet: face consistency is the minimum acceptable quality bar for a companion you're supposed to form a relationship with. the anchor cost is baked into the daily photo limits (5 on closer, 12 on bonded) to keep unit economics healthy.
the engineering detail: we author one reference face per companion using an authoring pipeline, save the pulid embedding, and every subsequent photo is rendered against it. outfits, poses, settings, lighting all vary — identity is invariant.
for users who want a specific look, you can also request things in chat: morning hair, coffee shop, a specific outfit. lucy's media orchestrator decides whether to send based on relationship stage, conversation context, and daily quota.
when you're ready to try: free tier gets 3 trial photos, no credit card. if she feels like the same person across all 3, you know the product is what it claims to be. that's the test.