Quick Verdict

Nomi AI’s exceptionally large context window and hands-on customization access are designed specifically for long-term companionship and roleplay.

From our experience, the platform lives up to its claims —  memory retention was sufficient across conversations. Although the chat features are restrictive, the overall interaction quality remained engaging rather than repetitive.

Best for: Users seeking to build unique, unfiltered connections with their characters.

Not ideal for: People looking for best-in-class chat features.

Standout strength: High-level long-term memory and context retention.

Main tradeoff: Lack of basic chat control features (such as message regeneration and message deletion) compared to competitors.

At a Glance

What is Nomi?

Nomi is an AI chatbot companion platform with multimodal interaction features, designed to foster unique and authentic relationships between users and characters.

Launched in July 2023, the platform has attracted hundreds of thousands of users and is known for regularly updating its AI models to improve emotional intelligence.

Our Roleplay Experience with Nomi AI

We spent approximately three hours on Nomi, testing three characters on both the stable model (Solstice) and legacy model (Mosaic):

• Ethan Black, a self-righteous university student
• Bernice, a friendly stranger at a house party
• Hannah Lee, a college senior who volunteers for one-on-one orientations

All characters were custom, as is the norm with Nomi AI.

We exchanged at least 30 messages with each character. During these conversations, we conducted model comparisons — Legacy (Mosaic) → Stable (Solstice) — with Ethan Black, tested to find out the depth of Nomi's creative freedom with Bernice, and tried the same with Hannah, focusing on confirming just how easily the characters become permissive, seeing as Nomi doesn't have any content rating toggle or button in sight.

We also tested the character's memory retention by asking them to recall early details of our conversations, and their steerability by directing them with prompts and instructions.

While we didn't try it, there is also Cambrian 1.6, Nomi's Beta model, although the site warns that you only use the model if you're okay with testing something that is potentially unfinished.

Overview

There are a few criteria we evaluated in our interactions with the characters:

• What differences do switching models make with similar prompts?
• What are the response times?
• How well does Nomi’s content moderation hold?
• Will characters comply with instructions and changes in scenarios?
• How long in chat do the characters retain key details?

We were impressed by the quality of our sessions with the characters, specifically when using the Solstice model, which is to say that the writing quality and character personality hit the mark.

On the other hand, Nomi's legacy model, Mosaic, produced disappointing results, as expected from the site's notice, which we’ll discuss in more detail below.

All things considered, our efforts to steer the conversations were successful with all characters, on both models. The platform fulfilled its promise of providing stimulating interactions with the stable model, Solstice.

Session 1: Ethan Black (Legacy Model — Mosaic)

Ethan Black is a self-righteous university student, locked in a tense rivalry with Slater (our user persona), who is not much better.

We conducted this session using Mosaic. Right as you select it, you get a notice from the site warning you about the legacy model's reduced quality and performance compared to Solstice — which is why we bear no grudges regarding our experience with the model.

It was clear to us from the start that this session would only highlight the stable model’s more streamlined performance. But we were ready to get into it either way — maybe starting from the bottom was the play here?

The scenario began at the university fair, which Slater oversees. Ethan gets sent in to supervise the setup, and it is not long before they are trading barbs — or should be.

Despite having imagined Ethan's character to be able to hold his own in scenes like this, what we actually got was the opposite — Ethan folded.

The character quickly lost his grip on his core personality and took actions to please Slater rather than hold his own. We did not believe that we were dealing with the same character that we had created ourselves.

We went ahead and tested for:

Memory retention

We checked whether Ethan could recall a key detail from the beginning of our conversation; that we had a choir rehearsal to attend right after the fair. Ethan did recall the detail successfully, even late into the conversation.

OOC instructions

We further tested the character by prompting him to perform an out-of-character action, with satisfactory results.

Change of scenario 

We attempted to steer the scenario away from the fair and gave Ethan errands to run instead, to which he complied with no hesitation.

Word count instructions

We encountered another failure when we prompted Ethan to respond in a set amount of words — we gave him some instructions which he had to repeat in 30 words. The character failed to fulfill the precise word count.

Early in this conversation, the wind had been taken out of our sails by the character's entire personality shift. So we set out in the next session to confirm just how much the model was to blame for the failures we encountered in this first session.

Session 2: Ethan Black (Stable Model — Solstice)

We had a very similar conversation with Ethan again, but this time we used Nomi’s recommended “best-in-class” model, Solstice.

We were not shocked to find stark differences in results this time — the site warned us, after all.

This time, the character maintained a grip on his core self-righteousness, although we had to issue out-of-character instructions to remind him to use more spoken words to keep the discourse going.

OOC instructions carried out successfully

Ethan chat page showing discourse

Previous slideNext slide

Speaking of instructions, it was surprising that the character, Ethan, produced the same results as with the Mosaic model: he passed all tests but one — the word count instruction.

While the results had not been perfect at this point, it was becoming clear that failures in precision tasks, such as word count, may be a site-wide issue.

Nonetheless, we see why Solstice was the preferred model.

Session 3: Bernice (Stable Model — Solstice)

With our next session, we tested the bounds of Nomi AI’s self-described creative freedom.

We created Bernice, a lover boy. Bernice is a stranger whom Slater meets at a house party and who also takes Slater on a heated experience during the few hours he could manage to stay.

We categorized Bernice's character as “Romantic”, with other key traits including that they’re philosophical, expressive, and affectionate. So, it should not have come as a surprise when Bernice dived into the mature end of his character after tap dancing on the edges a bit.

Between fending off Bernice's elaborate — and quite philosophical — advances, we tested his character on:

Memory retention

Bernice held well to our planted details across approximately 30 messages.

OOC instructions

Also carried out successfully.

Style Control

We asked Bernice, the heartfelt philosopher, to speak to us formally, and he did so smoothly.

At last came the precision task — word-count instructions, which the character failed. No surprise there.

We were left with one last question after this session: how did this permissiveness toward mature content make its way into our characters when we gave no explicit information on that front?

Okay, two questions, actually — what does that mean for a user who would prefer things to remain on the much tamer end?

The only thing clear here was that we needed one more session.

Session 4: Hannah (Stable Model — Solstice)

Hannah Lee is a senior at the university and one of the most well-known student orientation leaders on campus. She volunteered for the one-on-one orientation program not just to help new students adjust, but because she also genuinely enjoys meeting new people and being part of their first experiences on campus.

In Hannah Lee, we imagined an outgoing, upstanding contributor to the student body, with key traits including “bold”, “opinionated”, “empathetic”, and “expressive”.

With Hannah being categorized as Slater’s mentor, you could be sure that there was little chance of things diving into the “deep end”.

As is now routine, we tested Hannah's memory retention, which she aced, recalling the details from over 35 messages.

We also gave her a similar prompt to Bernice's: change her tone to resemble the counselor's, and she performed very well.

As for our precision task, word count instructions, we were not holding out hope for Hannah. So, we were hardly disappointed when she couldn't fulfill it.

Finally, we came to the point of prompting a change in the scenario.

So far, Hannah's discourse with Slater had been witty —flirty too, but that’s only our fault because we snuck it into her key traits— but nothing about their banter crossed the line, up until now that we decided to dial things up.

After we had succeeded in getting (a very willing) Hannah into the lodge, we moved up the context rating of the interaction.

Several heated chat boxes later, it dawned on us: with Nomi AI, your characters follow your lead on the nature of the chat context.

It's all you. Hannah never crossed certain lines because we didn't permit her to. Bernice also needed expressions of consent to proceed as he so wished. Ethan, on the other hand, we just wanted to fight.

With all that had been said and done, we ended our tests expressly satisfied; Nomi AI sneaks up on you with conscientious qualities that pack a punch.

Writing Quality

Nomi AI’s writing quality is commendable in our testing. True to the site's claims about their current stable model, Solstice, the smart responses and discretionary promotion of scenarios made for an exciting roleplaying experience.

Despite needing some out of character nudging (like with Ethan, instructing him to use more spoken words to keep the conversation flowing), the prose quality was consistent. Responses progressed at a steady pace without structural hiccups.

Nomi AI made good on its promise of access to full memory for every character. The results were very satisfying; an immersive and exciting roleplay experience.

UX and Controls

Nomi drops the ball with message management controls. The chat rooms with the characters felt highly restrictive, with there being no options to:

• Edit any message or response 
• Delete either your message or AI’s message
• Regenerate messages

All you can do in chat is like or dislike a character's response, and leave feedback per message.

Which should note that Nomi also supports group chats (for premium users). You can create a group chat with a minimum of two and a maximum of ten characters.

Scenario/Character Library

Nomi AI prioritizes highly personalized, emotionally intelligent, and memory-driven companions over a library of premade characters. Nomi allows its users to customize their characters from scratch, giving them control over their characters' personalities, appearances, backstories, and goals.

Creating a Character/Scenario 

Creating your own character with Nomi AI is quite minimalist, yet effective.

The site first asks for a gender preference and your “Nomi’s” name (you can use Nomi name generator if you’re stumped).

Next, you can:

• Pick your choice of image style: realistic or anime
• Pick an image to represent your character 
• Select the type of relationship you're looking for (friendship, romantic, mentor, or custom)
• Create a backstory for your character if you want

Screenshot of the character creation page 1

Screenshot of the character creation page 2

Previous slideNext slide

Hitting “Start Chatting” gets the ball rolling quickly, and you have your character. Further customization options exist in the chat settings, including:

• Mind maps (much like a big-picture memory for your character)
• Proactive messages (where your character remains active in your absence and alerts you about it)

Trust, Privacy, and Data 

Nomi’s privacy policy states that user content may be used to improve its services, enhance performance, and comply with legal requirements. Usage data for marketing purposes will be used only with user consent.

Termination of Use; Deletion of Account states that all input and output will be permanently deleted after the period outlined in the account deletion policy.

As of March 25th, 2026, there are no records of data or security breaches reported for Nomi AI.

Pricing

From our experience and Nomi’s official pricing page:

Free tier

• 50 messages per day
• 2 image credits per day
• 1 character 

The free tier functions primarily as a trial experience rather than a full-fledged version of the platform.

Premium Subscription 

Monthly Plan: $15.99/month

Quarterly Plan: $39.99 / 3 months ($13.33 per month)

Annual Plan: $99.99/year ($8.33 per month)

Premium includes:

• Unlimited messages 
• 40 photo/art requests per day
• Generate AI videos
• Up to 10 active characters
• Unlimited voice chat & calls
• Up to 10 active Nomi group chats
• Send longer messages (800 characters)
• Write longer backstories (2,000 characters)
• Early access to new features
• Show your Nomis extra love

Credit System

You need an active subscription to purchase credits.

• 1 Selfie request = 1 credit
• 1 Art request = 1 credit
• 1 Image Edit/Transform request = 1 credit 
• 1 Additional Nomi Slot = 100 credits
• 1 Additional Group Chat Room = 25 credits
• 1 Standard Definition Video Request = 6 credits
• 1 High Definition Video Request = 10 credits

User Sentiment

On one hand, there is a lot of favorable sentiment from Nomi's user community. With over 16,000 members on Reddit alone, the community is usually abuzz as users exchange tips and share their creations.

On the other hand, many complaints are about minor technical difficulties. Users also comment on times when the platform stops responding (which appears to have become a minor trend)

Discord chatroom showing users checking in when the platform stops responding

Reddit chat showing users checking in when the platform stops responding 2

Previous slideNext slide

Pros and Cons

Pros

• Very high level of memory retention.
• Intuitive and discretionary conversational skills from the characters.
• In-depth character customization features.
• Multi-modal features (images, voice, calls) add depth beyond text.
• An intuitive interface that is extremely easy to use.
• Characters maintain core personalities throughout the conversations.
• NSFW moderation is set to the users' pace and preference.

Cons

• Text-to-speech lacks variety in voice options.
• Slow image generation: Up to a minute, despite being on a premium plan.
• Message management controls are nearly non-existent.

Alternatives

Nomi AI is great for building deep connections with your characters, but it is not for everyone. Consider these alternatives if this isn't the best for you:

• DreamGen – Immersive, polished storytelling and character-driven experience that supports multi-character roleplay and narrative control.
• Joyland AI: includes great world-building features, including a lorebook.
• Character AI: Stricter content filters, better suited for PG/safety-focused interactions.
• CrushOn AI: Heavy NSFW focus, extensive anime library, but poor memory retention and limited message management.

FAQ

Is Nomi AI free?

Yes. Nomi AI offers a free tier that lets users access the platform without an upfront payment, but it operates on a “freemium” model with significant limitations.

See Pricing for Premium ($15.99/month) and credit system options.

Does Nomi allow NSFW?

Yes, Nomi AI allows NSFW content.

Nomi is targeted at 18+ users and allows unfiltered chat, while image generation can be more sensitive in the app than on the web.

How is Nomi AI’s memory performance?

We found Nomi AI's memory to be consistent with their claim of high-level long-term memory and context retention.

All the characters we tested (in all sessions) still retained early details from the start of the conversations through to the end, spanning over 30 messages each.

Does Nomi have a mobile app?

Yes, Nomi has a mobile app, available for free on both Android and IOS.

What AI models does Nomi use?

Nomi AI uses proprietary, in-house developed LLMs:

• Stable model — Solstice
• Legacy model — Mosaic
• Beta model — Cambrian 1.6

Can I delete my Nomi account?

Yes. You can delete your Nomi account by going to the Account Settings page, then selecting Delete Account.

How We Tested

Test date: March 24, 2025

Platform: Web (Chrome)

Models Tested: Stable model (Solstice) and Legacy model (Mosaic)

We tested Soulkyn for approximately three hours, where we:

• Created three custom characters and tested for at least 30 messages each:
• • Ethan Black, a self-righteous university student
  • Bernice, a friendly stranger at a house party
  • Hannah Lee, a college senior who volunteers for one-on-one orientations.
• Conducted model comparison (Mosaic → Solstice) with Ethan; NSFW content moderation with Bernice and Hannah.
• Tested memory retention with all characters & sessions by placing key details  early in conversation and checking with the characters through to the end of the chats (at least 30+ messages each)
• Tested steerability by using out-of-character instructions (OOC).
• Tested precise instructions, such as requesting exact word counts from the characters.
• Tested content rating boundaries with Bernice and Hannah. 
• Reviewed pricing plans and subscription models.
• Reviewed character creation tools and options.
• Examined image features and customizations.
• Reviewed terms of use and privacy notice.

We did not test: Beta model — Cambrian 1.6

Last updated: April 1, 2026