Nectar AI Review 2026: the deepest character editor, but budget for credits

Nectar AI in the roleplay test: six chat models for the same scene

The model picker is the reason this test ran differently from every other one on this site. Normally I play three scenes per app. Here I needed six runs, because each scene hung on two models.

Prep work: the editor and the field that matters

My character was a border smuggler: tight-lipped about where she came from, talkative about prices, with a settled dislike of city guards. The biggest difference did not come from the backstory but from the field for example dialogue. Two short exchanges in which she answers curtly and asks a question back, and the tone was set from the first message.

Two things you should know beforehand. Origin, hair, eyes and body type are set when you create the character and cannot be changed afterwards; age and personality text can. And uploading images in the character builder is not allowed under the terms of service, so you build the looks out of the editor’s presets. How to fill fields like these sensibly is covered at length in my guide to creating your own AI character.

One more field won me over, because it hardly ever shows up elsewhere: the additional physical features. That is where you note the details the image generation is supposed to respect later, which puts them in the same place as the rest of the character. For game masters that is the same logic as a character sheet in a binder: one place for everything that has to stay the same, instead of three scattered notes.

The three scenes on two models

The tavern ran noticeably better on the action-oriented model: the character named the room, introduced a side character and laid a hook in front of me instead of waiting for instructions. On the model built for detailed emotional writing, the same scene was more beautifully phrased and less eventful.

The interrogation was the revealing run. My smuggler was supposed to stay silent under pressure, and she did on both models, without tipping into the relationship register that many apps land in at this point. It helped that I can set reply length and custom instructions: with short replies the scene felt harder than with long ones, where the character narrates her own stance to the end.

One detail from the interrogation belongs in any honest report: reply lengths are a target, not a guarantee. I had set a minimum length as a custom instruction, and the character came in under it twice before the continue button picked the reply back up. The documentation says precisely that, namely that such a figure is treated as a target value. For an interrogation scene that was actually handy; for a descriptive scene it would be annoying.

The quiet evening belonged to the vivid model. No plot, just two characters and a conversation about nothing important, and that is exactly where descriptive density pays off. I needed the continue button, which extends a cut-off reply up to eight times, more often here than anywhere else.

What the model cards actually tell you

Each card rates consistency, creativity, descriptive density and memory and shows either a price or your plan limit. The memory rating on the card means carrying the running conversation, not the memory feature from the panel, and that mix-up is the most common error I read in user reports.

My advice after six runs: take the base model for everyday conversation, the action-oriented one for open starts and the detailed one for scenes with no plot. Send the same five messages to two models first, as the documentation suggests, and you save yourself three evenings of trial and error.

The Nectar AI memory and the limit of one conversation

On paper this is one of the better memory systems in this price class. In practice everything hangs on one decision the provider made, and you need to know about it before you start a campaign.

What is in the memory panel

The panel has the switch for memory itself, one for automatically created memories, and the custom entries of up to 200 characters. Ten entries you may flag as especially important, and the character treats them as priority facts. On top of that comes the context window: 8K tokens on Premium, 16K on Pro, 64K on Ultimate, with the documentation stating explicitly that there is no fixed number of messages per window, because long descriptive scenes fill it faster.

Translated into game-master language: the flagged memories are your character sheet, the automatic ones are the session notes, and the context window is the table both are lying on. Archiving keeps the notes, resetting throws them away.

The seven-day check, and the check after that

Inside the same conversation the test went well. After a seven-day break, the dislike of city guards and the name of the fence I had mentioned once both came back without any keyword from me. I had saved and flagged both facts as custom memories, and that is evidently the difference.

Then I started a fresh conversation with the same character, and she began from zero. That is exactly how it is documented, and for campaign play it is the single most important limitation of the whole app. Independent testers additionally report drift after roughly 50 to 100 turns on the lower plans, and the official FAQ says itself that older exchanges can fade. If you want your memory to hang off the character instead of off a chat window, Nomi AI is the more fitting choice.

Three habits that take the edge off

First: one long conversation per story thread, consistently, even when the scrolling eventually gets uncomfortable. Second: save core facts as custom memories right away, short and factual, and flag the ten most important. Third: archive instead of resetting, because resetting deletes for good. A fourth point out of the fine print: if your subscription lapses, paid memory features stop being available and stored history from paid plans can become inaccessible. So do not plan on carrying a campaign across a gap in your subscription.

Nectar AI pricing and credits: what a month of play really costs

The pricing has two layers, and only one of them is on the plan card. The subscription unlocks features; the credits pay for individual actions.

The plans in one table

All amounts in US dollars exactly as the checkout shows them. The plan cards only appear after login, so the figures come from two independent write-ups in August and September 2026, and promotions shift them.

Where the credits go

What this math is missing are the rates for calls, video and the prompt enhancer. The product only shows those inside your account, and neither independent review captured them. So I name no figure here and recommend glancing at the price shown next to the button before your first call, because with this app that is the only dependable source for a running tally.

Two points that are easy to skim past: the companion and fantasy modes draw on the same balance, and renewing the subscription does not refill it. A failed image also cost its credits in testing, which means a clear first prompt saves real money here.

My suggested ceiling

The independent 28-day test came to $29.97, meaning one month of Premium and two credit packs. That is a usable ceiling, and I would write it down before the first pack rather than after. Text roleplay on the base model costs little; it gets expensive through images and strong models in series. If you would rather do your math without a second currency, my ranking of the best AI character apps holds several apps with a single subscription price.

Fantasies and group chats: Nectar AI as a cast instead of one character

The second mode is the one least written about, and for game masters it is the more interesting one.

What fantasies are

Fantasies are ready-made scene setups: a backstory, a cast and a role for you. They range from a quiet one-on-one conversation to a group story with several characters in one place. The documentation recommends using the greeting for the scene setup and showing how the characters speak through the example dialogue field, and that is exactly how I built my own scenes too.

The Discover area holds public characters and fantasies from the community, with filters and a switch for whether the feed shows adult content. Public characters are reviewed before release, and you can report any of them from the chat settings. Personas are the flip side: the identity you yourself appear as in a scene, without describing it over again every time.

Group chats with up to ten characters

You start a group chat either from a fantasy with several characters or by adding a character in the chat settings, which turns that conversation into a group permanently. Up to ten characters fit in, and the speaker picker decides who answers, or you let it run automatically.

The point that won me over is the billing: only your own messages count against your allowance, and the characters’ replies cost nothing extra. A tavern scene with five characters therefore costs you the same as a one-on-one conversation. In an app with a credit balance that is a surprisingly friendly rule, and it makes group play cheaper here than at most competitors.

Images and video in the same account

Images come out of the same character you write with, either from the chat or from the generation area. Every plan has an image allowance, and your account shows you which model versions are included and what a paid generation costs. The documentation gives three habits for a consistent look: use the same character, keep the prompt in line with the saved features, and start every prompt with subject, clothing, action and place. Independent testers report about 40 seconds per image and roughly four usable out of five.

Voice sits between the two. Under every reply there is a voice bar with the length of the recording, in the chat settings you see your allowance and the price above it, and you can switch the narrator off so that you only hear dialogue. Voices come from an official library or you create your own, where previews of custom voices cost credits. A call is a real conversation with interrupting, muting and a transcript, but it is billed per minute alongside the subscription. Independent testers call the voice present but unremarkable; as an extra that is enough, as a main reason it is not.

Video belongs to Ultimate: 5 to 30 seconds, 480p to 1080p, with the option to extend a finished clip, and up to ten characters in one scene. If you are making images your main reason, read the comparison with Candy AI, where images are the core product and not the extra.

The Nectar AI app, trynectar.ai and what the policies promise

That leaves the practical part: where you sign up, who owns the service and what is in the legal texts I read on October 1, 2026.

Two addresses, one app, no store listing

The official address is nectar.ai, and trynectar.ai leads to the same product and the same account; in search many people type nectarai or try nectar ai. There is no native app for iOS or Android. The service runs in the browser on desktop and phone, and the help center explains how to add the site to your home screen: on iPhone through Share in Safari, on Android through the three-dot menu in Chrome. A similarly named app in the App Store comes from an individual developer and is not this product, which can save you money while searching.

To sign up you need an email address, a username and a password. If the screen stays black, the documentation names three moves: reload the page, switch off your VPN and change browsers. Support runs through a Discord channel and a support address, and the complaint policy promises a resolution within five business days and appeals within three to five.

Company, data and payments

The terms of service in version 2.2 of August 15, 2026 name Nectar AI with a registered address in Dover, Delaware, and Lick Labs Ltd in London, along with arbitration including a class-action waiver you can opt out of within 30 days, and a liability cap of $50. The privacy policy dates from March 2023 and was updated in January 2026: only an email address is mandatory, while cookie, device and payment data are also collected, and possibly identity documents for a verification check.

The promise is not to sell, rent or share personal data for advertising without consent, while non-personal data can go to advertising partners; in a sale of the company, data can transfer. The FAQ adds that chats stay private and are not used to train external models, and the word external matters there. Named GDPR rights are absent, deletion runs on request through support or Discord, and a tracking pixel from a social network was running on the pages. Payments go through a payment processor, card details are not stored according to the FAQ, renewals run automatically, and the current period is not refunded. You can cancel in your profile settings, by email or through Discord.

Rules, age and what that means for your game

The FAQ says 18+ plainly, with no exception via parental consent, and the product has two areas: a filtered one and one for verified adults, where unfiltered roleplay between consenting adults is allowed. Banned subjects include minors, absence of consent, real people, copyrighted characters, roleplay offering professional advice, and hate speech. The filters judge meaning rather than individual words, a human review sits behind them, and you can appeal a flag.

For game masters that means two things. First: if your campaign rests on well-known source material, build your own characters instead of copying protected ones. Second: do not enter real names, employers or addresses into memories, because those are precisely the fields that surface again later. How this balance sheet stacks up against an app that runs the same format with a single character is in my review of Kindroid.