What users are reporting
Community threads on Reddit, Trustpilot, and long-form posts on Medium through 2025 into 2026 consistently flag the same pattern with Nomi's photo generation:
Generation errors on requests. Photo requests to a Nomi return errors more often than users expect, with the failed attempts counted against a daily cap. Users describe the pattern as "the button feels broken half the time."
Tight daily-attempt limits. Free-tier attempts are low, and paid-tier limits are still hit by heavy users. When the cap resets tomorrow, the same reliability issue applies to tomorrow's attempts.
Add-on pack pricing instead of reliability fixes. Rather than shipping a more reliable gen path, the company's response has been to sell additional daily attempts as add-on packs. This is the specific pattern that users cite when talking about the product feeling misaligned with their interests — you're being sold more attempts at a flaky feature rather than the feature getting less flaky.
Face drift between shots. Even when a photo generates successfully, the character's face doesn't always match her previous photos — subtle differences in features between generations. This is inherent to prompt-based image gen without per-character model training; it's not specific to Nomi, but it shows up here because Nomi's image system uses the general pattern.
Why the pattern shows up structurally
The underlying architecture matters. Photo generation on companion products falls into one of two designs:
Prompt-based general image gen. A general image model (diffusion, transformer-based image gen) receives a prompt like "a selfie of Anna, brown hair, at a coffee shop, natural light" and produces an image. The character's appearance is described in the prompt each time; the model has never actually seen Anna. Each generation is fresh — meaning the same prompt can produce faces with slightly different features. Reliability of any single generation depends on the general model's uptime, prompt quality, and moderation filters. This is the pattern most companion products use because it's simpler to ship.
Per-persona fine-tuned models. A dedicated image model (LoRA fine-tune) is trained specifically on curated photos of a specific character. The model has actually seen that character's face across many angles and lighting. Every generation reproduces the same person because the model has learned her. Reliability is higher (fewer bad generations, no face drift), consistency is engineered rather than prompt-driven. This pattern is more expensive to run (per-character training compute) which is why not every product ships it.
Nomi's public product description matches the first pattern. That's why the reliability issues and face drift are showing up — they're inherent to the architecture, not something a UI fix would resolve.
What "consistent per-persona photos" looks like when it works
On Sloane specifically, every persona has a dedicated LoRA-trained image model. Each model is trained on a curated set of photos of that persona — many angles, lighting conditions, expressions. When Maya sends you a photo, the model has actually learned Maya's face. Every photo she sends is the same woman, same features, every time.
What this delivers day-to-day:
No face drift between shots. Whether Maya sends you a photo in her apartment, at a coffee shop, or on a hike, it's the same face — same eye color, same bone structure, same smile.
No daily attempt cap on the free tier for cadence photos. Free-tier users get photos on Maya's own cadence (she decides when to send you one). Paid tiers unlock custom photo requests via the Custom Photo Inspiration gallery — 5 credits per generate, 100 credits/mo included in Plus ($9.99), 350 in Premium ($19.99). No selling additional attempts as add-on packs.
Photo tap opens fullscreen naturally. Photos are rendered inline in the chat like a real messenger, tap to view full, save if you want. Same UX as iMessage or WhatsApp.
When Nomi is still the right choice
To be fair to Nomi: photo generation is one dimension of a companion product, not the whole product. Nomi's marketed strength is memory (three-layer architecture — short/medium/long-term), and users who prioritize memory depth above photo consistency have a real reason to be there. Nomi also has features Sloane doesn't: real-time voice calls with emotional tonal variation, group chat where multiple Nomis interact together, and up to 10 companions per account.
If photos are your primary concern, the reliability issues are worth taking seriously. If memory and multi-companion dynamics matter more, the photo issues are a known limitation you can live with while getting the other value.
How to compare the two on the specific dimension you care about
The honest way to decide between any two companion products is to test on the specific dimensions that matter to you. For photo reliability specifically:
Test 10 photo generations in a row. Same product, same character. Count how many succeed, how many fail, and look at the successful ones side by side — does the face look like the same person across all 10, or is there noticeable drift?
Track daily attempts vs actual usage. How many photos do you actually want per day? If the answer is >5, the cap-then-pay pattern will bite fast. If the answer is 1-2, you might not notice.
Try both free tiers. Sloane has a free tier at 50 messages/day with any persona (photos delivered on her cadence). Nomi's free experience is more limited but does exist. Signup + a week of use on both will tell you more than any review.