Lensa, FaceApp, Remini — and where Precog stands inside the AI-self wave
A short cultural history of the AI portrait apps that came before us, what each got right, what each got wrong, and the small corner of the territory we are trying to occupy.
The cultural appetite for seeing yourself, processed by AI has produced a very specific shelf of consumer products over the last six years. Precog sits on that shelf. We are not the first product to put a generative model in front of a user's selfie, and we will not be the last. This essay is a short history of who came before us, what they did, and the narrow corner of the territory we are trying to claim.
We will be honest about the comparisons. None of these products are bad — most of them are technically impressive. They simply answer different questions than the question Precog is trying to answer.
The wave
A short timeline of the AI-self products people actually used:
- 2017 — FaceApp. The first mass-market AI age-progression. Showed you yourself older. Optionally younger. Optionally a different gender. Built primarily for shareable shock value. Brief privacy controversy in 2019 about Russian-affiliated cloud storage.
- 2022 — Lensa AI ("Magic Avatars"). Not aging — style transfer. You upload 10–20 selfies; the app spits out 50–100 stylized portraits (anime, fantasy, royalty, sci-fi, etc.). Brief sensation. Ethical controversy about training-data sourcing.
- 2023 — Remini and the "deep restoration" cohort. Old-photo restoration, plus increasingly polished selfie enhancement. Less about transformation, more about "make this image look better."
- 2024 — Loose category of identity-AI apps. Linkedin headshot generators (HeadShotPro etc.), age-progression toys, dating-app photo generators, fitness "what would you look like with 8% body fat" tools.
Each of these answered the same surface question — what does AI think of my face? — but with very different underlying purposes.
What each was actually for
FaceApp was for sharing. The aging filter was the viral move — share the surprise, show your friends, post on social. Engagement loops were optimized for novelty. After two or three sessions, most users were done.
Lensa was for self-image. The avatars were stylized in ways that were flattering across a broad set of styles. Users, especially in late 2022 when the trend peaked, used them as profile photos, social posts, dating-app pictures. Lensa knew what it was doing — make me look more interesting than I look.
Remini and the restoration cohort were for un-flattening time. Old, blurry, damaged photos restored. Family ancestry images sharpened. The use case was different: not "what could I look like" but "what did this person actually look like, recovered from photographic decay."
Linkedin headshot apps were for utility. Cheap professional photos, no studio booking, good enough for a profile. Pure utility.
The common thread is that all four of these answer some version of make me look different, with different purposes. Sharing. Self-image. Recovery. Utility.
Where Precog is different
Precog does not make you look different. The portrait is not aging-for-shock. It is not stylized. It is not a dating-app upgrade. It is not a Linkedin headshot.
It is, instead, a model of where your current behavior is taking you.
The technical similarity is real — a generative image model takes your selfie and produces a transformed version. The product is fundamentally not the same. The transformation is driven by behavioral data — your sleep, your exercise, your hydration over a specific week — and the output is meant to be informational, not aesthetic.
This changes everything downstream:
| Aspect | FaceApp / Lensa | Precog |
|---|---|---|
| Input | Just the selfie | Selfie + week of habit logs |
| Output frame | Static, immediate, one-time | Recurring, weekly, indefinite |
| Purpose | Share, decorate, restore | Inform, calibrate, decide |
| Engagement loop | Novelty (used 2–3 times) | Ritual (used 52× / year) |
| Output value | Aesthetic (better-looking you) | Calibration (more-honest you) |
We say this not to claim superiority. The AI selfie market is broad enough for many shapes. We say it because the most common misunderstanding people have when they first hear about Precog is "oh, like FaceApp aging?" It is not.
What the previous products got right
We learned from each of them. The honest credits:
- FaceApp's identity-preservation — the engineering work that keeps the aged face recognizably you is genuinely hard, and our use of it would not be possible without the lineage of work FaceApp's team kicked off.
- Lensa's onboarding — Lensa nailed the brief moment of "upload your photos, here is the output" with very little friction. Modern AI-selfie products that take longer than three minutes to first output lose users.
- Remini's quality bar — the cleanest restoration model in the consumer space sets the visual quality expectation. Precog's portraits have to clear that bar.
- HeadShotPro and friends' demonstration that there is consumer demand for AI imagery of yourself that is not for entertainment — for utility. Precog is in a different genre but in the same neighborhood.
What they got wrong (or didn't try to do)
The single thing none of them did was cadence. Each was a one-time or few-time product. You used FaceApp for a week in 2017 and you were done. Lensa was a single afternoon's project. Remini you returned to when you had old photos to restore.
None of these built around the idea that the AI image of yourself could be a recurring informational artifact — generated weekly, drawing from your real behavior, accumulated as a longitudinal record.
That is the corner of the territory Precog is trying to occupy. Whether it works as a consumer product is a separate question. The argument for why this corner is worth occupying is the foundational essay set on this site (start with the Stanford 2011 study and the Hershfield framework).
A short note on competitors who haven't shipped yet
There will be more entrants in this space. The technical components — generative image models, identity preservation, on-device or cheap cloud inference — are commoditizing fast. By 2027 the question of "build a recurring AI portrait product" will not be a moat question; it will be a habit-and-distribution question.
We are not pretending we are the only people who could build Precog. We are betting that the first product that takes the recurring weekly cadence + behavioral input combination seriously will be the one that finds the audience for it. We are trying to be that product.
Where this leaves the user
If you are someone who has used FaceApp or Lensa and felt the small high of "look at AI me," we understand the appetite. Precog is not that product. The portraits are quieter. The engagement loop is slower. The point is not the surprise; the point is the calibration.
Some people will find that a relief. Others will find it boring compared to the pop-style of the entertainment products. Both reactions are valid. We are trying to build for the first kind of user.
— Codeful