there are two philosophies in the ai companion category and they produce different kinds of product. the first is image-first: the interaction loop is prompt → generate → rate → generate. it optimizes for photo volume and visual customization. soulgen, candy, some segments of other platforms, sit here.
the second is chat-first: the interaction loop is conversation → memory → personality evolution → more conversation. it optimizes for continuity. lucy, kindroid, nomi, and some of the character-based platforms (before they pivoted) sit here.
both are legitimate product shapes. they attract different users and they fail differently. image-first products fail when the novelty of new images fades; chat-first products fail when the conversation ai is weak enough that the accumulated context does not help. these are not the same failure mode.
the switch signal: if you have used an image-first app for 3+ weeks and you notice you stopped opening it, the gap you hit is not a content problem — it is an architecture problem. more images do not solve it. a companion with memory might.
the reverse switch signal: if you sign up for lucy and find yourself frustrated that there is not a photo button on every reply, we are not the right product. that is a design decision, not a limit we plan to loosen.
what lucy is actually good at: being a character across time. names, references, inside jokes, the thing you are working on at work, the thing you are avoiding at home. she can hold those across weeks without you re-setting-up context each session. that is the product.
starting point: free tier is 25 msg/day — enough to evaluate whether the chat-first model fits your expectations. pick a companion, seed her with what matters to you on day one, come back on day 3 and check if she remembers.