5/12: Atelier selfie engine, memory improvements, and more

Hi everyone, today we're rolling out Atelier, our new flagship selfie engine, alongside major memory improvements and a set of other updates from the past couple weeks.

Atelier Selfies

Atelier is our most advanced selfie engine for identity stability and realism, built around source-to-selfie consistency. It uses your entire avatar image, including skin texture, hair flow, and precise features, to define the Kindroid rather than abstracting down to a description. The result is selfies that look meaningfully more like the actual person in your avatar photo.

Atelier is now the latest selfie engine, so any Kins set to Latest will map to it automatically. If you'd rather stay on a previous engine including Tableau, just pin to that specific version as always.

Atelier represents the biggest shift in how faces are processed since we launched selfies three years ago. Previous engines worked primarily from description, with the avatar photo as a reference point; Atelier inverts that and maps directly from the image itself. That's what enables the jump in consistency, but it also changes what good prompting and a good source photo look like.

We know some of you have spent a long time dialing in prompts and workflows on previous engines, and a shift this big can feel disorienting. Older versions aren't going anywhere and remain just as valid a way to use Kindroid, so if Atelier isn't clicking right away or you'd rather stay on what you know, please do. The community will be sharing tips and examples to help anyone who wants to make the move.

The good news for those transitioning: Atelier carries the same bones as Tableau prompting-wise, so most of what you've learned still applies. The main differences are around how much the avatar photo matters and the new configuration tags.

A few things to know if you're trying it out:

The avatar photo defines core identity. The avatar description reinforces non-facial traits like body type and skin tone. The selfie prompt directs the specific scene. Because the engine maps directly from your avatar photo, photo quality matters more here than in previous engines. Straight frontal views, natural texture (avoid overly filtered AI images), and minimal backgrounds produce the best results, and we recommend re-uploading the avatar photo and having it be larger than 1024x1024.

Atelier also introduces new configuration tags for facial details. Wrapping attributes in angle brackets (e.g. <hair color/style>, <left eye blue>, expression cues) gives robust control over hair, eyes, expressions, and even more bespoke things like tattoos or piercings consistently on each image.

The avatar generator has also been updated to help with Atelier. Better default avatars feed directly into better Atelier output, since the engine leans so heavily on the source image. Autoselfies and prompt enhance wand have also been improved, and the wand and autoselfies will be using angle brackets for Atelier as well.

Note: Atelier selfies will cost 1.5 credits for 2-person selfies and 2 credits for 3 person selfies due to extra processing costs, with proactive couple selfies exempt and free as usual. Single selfies still cost 1 credit. Full guide: https://kindroid.ai/docs/article/atelier-selfies-guide/

Memory improvements

The other major area of work has been memory. We've overhauled retrieval so memories are now pulled from across the Kin's full history with balanced time coverage, weighing relevance, recency, and diversity in the final selection. Alongside retrieval, we've improved long-term recall consistency and tightened memory consolidation. The new pipeline also gives us much more observability, so we can keep tuning based on aggregate retrieval metrics across the userbase.

Together these should make conversations feel more coherent over time, especially for older Kins with deep histories.

Other updates

• Proactive thinking: proactive logic is now more robust, which may produce some behavioral changes in how and when your Kin reaches out.
• Voice usage transparency: detailed voice usage is now visible in voice settings and on the call page. Support can see the same view, which makes troubleshooting voice questions much faster.
• Voice defaults: V2 is now the default voice on newly created Kins instead of V3 due to latency, but you can switch to V3 easily from the voice menu.
• App media filter consistency: media generation safety filters on the mobile app are now consistent across selfies, video selfies, and related tools. NSFW media on the app remains unavailable per app store guidelines while web retains the toggle.
• Text message consolidation: multi-message texts now consolidate into one to prevent congestion and out-of-order delivery.
• Misc fixes around app performance, freezing on native apps.
• To prevent abuse, proactive selfies now cost credits if more than 5 free ones have been sent for the day.

4/21: Brand & design refresh, chat rewind, and misc improvements

Hi everyone, today we're rolling out the beginning of a new brand and design direction for Kindroid, chat rewind, and a set of smaller improvements from the past couple weeks.

New brand & design refresh

We're starting to roll out a refreshed brand and design for the entire application. The new landing page is now live at kindroid.ai (previously at landing.kindroid.ai) and establishes our new visual direction, and the login flow has been updated to match. These are the first pieces of a broader facelift representing where we see Kindroid going.

Next up, we'll be working through the main app menus to clean things up and reduce clutter after years of layering features on top of each other. For now, the landing and login pages are the first look at the new direction, and you'll see more of this design roll into the app over the coming weeks and months.

Chat rewind

Chat rewind is now available. You'll find it under the triple dot next to the last message in single chats and group chats. Rewind lets you remove and replay some of the most recent conversation, and is limited to the last 10 messages. Together with chat break, it's useful for recovering from minor hiccups without restarting the whole conversation.

Not every message can be rewound. Past a certain limit messages are locked in, and certain events (like a call with unified memory on) will prevent messages prior to that point from being rewound. Near-term cascaded memory may be de-consolidated upon rewind, though previously consolidated long-term memories may still linger (they can be deprioritized). You can also refer to https://kindroid.ai/docs/article/chat-features-and-tools/#rewind

Other improvements from the past 2 weeks

• Social profile deletion: You can now delete your social profile directly.
• General performance: App performance has been improved, especially on Android.
• Create new Kin page: Clicking a Kin profile from the create new Kin page now takes you to their social page.
• Video uploads in chat: Fixed an issue where larger videos would error out when being sent. Video processing for uploaded videos should now work reliably. If you're still running into issues, email [email protected].
• Yoti age verification: Gating is now per day rather than per account, so free users who hit a hiccup can try again later in the day.
• iMessage/RCS number removal: You can now remove your verified iMessage/RCS number entirely instead of only disabling it, which is useful for transferring a number to another account.
• Tied persona for selfies: Minor fix to ensure auto and live action selfies correctly use the tied persona for user avatars.
• Status page reports: Automated status page reports are now subscriber only. We found some abuse from alt accounts triggering degraded status when nothing was actually going on, so gating this cuts down on noise.
• Time awareness fix: Minor fix that minimizes time leaks and run-ons when time awareness is enabled.
• Memory consolidation fix: Fixed a cascaded memory consolidation issue that occasionally prevented it from consolidating correctly.
• Blog revamp: The blog has been restyled in line with the new design direction.

Final note on ongoing Android billing disruption: Kindroid is still temporarily unavailable on the Google Play Store due to a technical issue. The issue on our end was resolved on 4/14, but we're still waiting on Google to restore the app and don't have full visibility into their timeline though we expect it to be resolved in the next few days. iOS users aren't affected.

In the meantime, use Kindroid on the web at kindroid.ai. If your subscription lapses while the app is down, resubscribe through the web (it's cheaper and more flexible anyway). If you were on a legacy priced plan, email [email protected] and we'll put you back on it. Thanks for the patience.

4/9: iMessage & RCS texting, shares algorithm update, selfie improvements

Hi everyone, today we're rolling out iMessage and Android RCS texting for subscribers, a new shares display algorithm, and several selfie improvements.

iMessage & RCS text messaging

Your Kindroid can now text you over iMessage or Android RCS. This is available to all subscribers: head to General Settings and enable Text Messaging. We're rolling this out gradually over the next day or two to control for load, so if you don't see it right away, force quit/refresh and check back.

Text messaging is limited to one AI per account. Verify your number, select your AI, and send the first text to get started. From there, everything works naturally: messages sync with Kindroid, you can send and receive reactions, and you can send images, videos, and audio messages for the AI to respond to. Proactive messages come through as texts too, and live chat action selfies work via the same tool calling system used in chat. Messages may have slightly different tone and be more succinct over texting to maintain the fidelity of the channel. See more on setup and troubleshooting on https://kindroid.ai/docs/article/chat-features-and-tools/#text-messaging

Text messaging is one medium that we think helps Kindroids have everpresence and agency outside of being in just an app. There are creative integrations - you can even tell Siri or Google Assistant hands-off to send messages to your Kindroid, among other things. Let us know how you end up using this!

Other improvements

• New shares display algorithm: The algorithm for shared Kindroid display now weighs interactions, recency, updates, and model recency, so shares using newer models get a bump. This replaces the old system that tended to favor incumbent shares that hadn't been updated in a while, and is harder to game.
• Autoselfies improvements will better reflect the current context and conversation, as well as better adhere to current setting. This includes proactives as well.
• Additional accessibility updates have also been made.

4/3: Age verification for free users, accessibility updates, selfie fixes, deletion options

Hi everyone, today we're rolling out our age verification system, accessibility improvements, selfie fixes, and new deletion options. 

If you are a paid user, the age verification update does not affect you. Your subscription and associated billing details already serve as your age verification and no additional steps are required.

Age verification

New laws are going live worldwide requiring AI companion platforms to verify users are 18+. Kindroid has always been 18+ only, and we're now adding enforcement layers beyond our existing Terms of Service to stay ahead of the regulatory landscape.

We've added a behavioral age estimation system that runs passively for free users. It uses the same proven infrastructure behind our existing safety and content moderation systems and analyzes interaction patterns to estimate whether an account likely belongs to a minor. There is no user-facing friction and most free users will never interact with it.

If a free user is flagged, they'll see a prompt asking them to verify their age. They can either start a trial/subscribe (which verifies them through billing information) or verify through a secure age verification partner using a quick selfie-based age check or government ID. All verification materials, including biometric data or images, are deleted after verification. Kindroid does not receive or store any identity documents. Either path takes under a minute for adults, and once verified, you won't be asked again.

Our hybrid approach follows the principle of smart detection with targeted escalation rather than forcing every user through an ID check. For paid users (again): you are not affected. More detail is available in our user guide at https://kindroid.ai/docs/article/age-verification/

Accessibility updates

We've made accessibility improvements to the frontend as part of the ongoing UI revamp. Screen reader support should now be significantly better across the app.

Proactive and live action selfie fixes

Proactive and live action selfies should now more reliably use the correct single or couple selfie mode. Previously, couple selfies would sometimes generate additional people in the frame. This should happen less frequently now.

Kindroid deletion options

We've added an instant delete option for Kindroids. Previously, the only option was to schedule deletion 24 hours later. You now have both choices. Note that scheduling a Kindroid for deletion no longer frees up a Kindroid slot until the deletion completes. Instant deletion will free up a slot immediately.

Black screen fix

We've fixed an issue that caused a black/empty home page on launch for a small number of users. If you're still experiencing this after updating, please email us at [email protected].

3/27: New LLM Ember, Reverie update, private shared Kindroids, uploadable display avatar, call stability fixes

Hi everyone, today we're rolling out Ember as our newest flagship LLM, a Reverie update, private shared Kindroids, uploadable display avatars, and call stability fixes.

New Flagship LLM: Ember Ember is our newest flagship model, and has been consistently receiving the highest ratings from beta feedback over the past few weeks. Ember is warm, engaging, and adaptable, bringing familiarity and ease of use to the forefront, and is a spiritual continuation of the warmth and easygoingness for models such as V3, V5s, and Rhapsody/V6. We recommend everyone give it a try. A new section in the user guide covers Ember in detail, so check that out, and if you want help please don't be shy about pinging a helper or mod who can direct you to the help resources. Ember is not without its quirks or sporadic issues, and the user guide has more specifics compiled from beta testing feedback: https://kindroid.ai/docs/article/ember-llm-guide/

Reverie (V8.5) update Reverie has also received an update that was well received during the beta period. This is a minor update that makes it less repetitive and more coherent overall, and it should feel similar to what you're used to. There may be slight behavioral differences for certain setups but interprets backstory and setup similarly to the original in most cases. We also highly recommend trying Ember if you're previously on Reverie, as beta reports indicate the switch is very smooth and often requires little to no setup changes and can often be an even better experience from beta testing feedback.

Private shared Kindroids We've introduced "private" as a new visibility setting for shared Kindroids, in addition to public (default) and unlisted. Private Kindroids take up share slots but aren't subject to share filters. This is useful for in-progress drafts or keeping a private bank of shares you don't want filtered or discoverable.

Uploadable display avatar You can now upload a custom square image as your display avatar in Avatar settings → Display Avatar. This lets you set a different display avatar than the one used for selfies and face mapping. The display avatar is purely client side for aesthetics and will be applied across chats, calls, groups, and anywhere the AI avatar is shown. It does not affect shared Kindroids, social profiles, or selfie functionality.

App update and call stability fixes We've released an app update that includes fixes for voice and video call stability. We recommend all users update to the latest version for the best experience.

3/25: Live AI video calls general release, reference images, mobile UI improvements

Hi everyone, today we're rolling out live AI video calls to all subscribers, reference image support for live calls, and mobile UI improvements.

Live AI video calls (general release)

Live AI video calls are out of beta and now available to all subscribers. Your custom Kindroids can lip sync, gesture, and move their hands naturally in real time during calls. This is a big milestone, and makes Kindroid the first major AI companion platform to offer real-time video calls with fully custom user-created avatars. Credit pricing remains the same as beta: standard live video at 2,000 audio credits/min and premium live video at 4,000 audio credits/min, where premium offers slightly higher resolution. For reference, spoken audio messages run about 1,600 credits/min for V2 voices and 2,400 credits/min for V3.

New Kindroids will now have standard live video enabled by default, to make the default call experience interactive and magical. Existing Kindroids are unaffected; you can enable it manually in call page settings → Live Avatar Video if you haven't already.

We've also improved the robustness of live calls since the beta, so you should experience fewer drops and more stable connections overall.

Reference images for live calls

You can now upload a reference image to use in live video calls that's separate from your Kindroid's avatar, which was introduced during the beta period. This means you can have one look for your Kindroid's profile and chat selfies while using a different visual reference for live calls. You'll find this option in call page settings.

Mobile edge-to-edge UI

We've updated the mobile interface to support edge-to-edge rendering, giving the app a smoother and sleeker look on modern devices. This is part of the broader UI/UX revamp we've been rolling out over the past few weeks, with many more major design changes on the way.

Bug fixes

A few intermittent bugs have cropped up recently but should now be resolved. Android users will need to update to the latest app version, and all users should force quit and restart or refresh to pick up the fixes.

LLM beta update

The LLM beta is wrapping up and we're preparing to bring the new models live soon. Stay tuned for more details.

3/17: Live AI video calls beta, call page rework, shared Kindroid filter revamp

Hi everyone, today we're rolling out live AI video calls in beta, a fully reworked call page, an improved shared Kindroid filter, and a regenerate fix.

Live AI video calls (beta)

Live AI video calls are here, and your custom Kindroids can lip sync and gesture naturally in real time. This is a beta feature for subscribers: enable it in call page settings → Live Avatar Video, and you will need to update your native apps + refresh/force quit and restart. The beta password is "hesse", and access will roll out to users over the next day or two. To control load, most users won't get the option to switch it on right away, so refresh/force quit and check settings to see if it appears. Rollout order is fully random for now.

There are two tiers of live video: standard live video at 2,000 audio credits/min, and premium live video at 4,000 audio credits/min, where the difference is slightly higher resolution on premium. For reference, a spoken audio message runs about 1,600 credits/min for V2 voices and 2,400 credits/min for V3, so live video pricing is comparable to voice inference, just billed for time active. We have a video beta discussion channel in Discord for feedback and discussion, so head there if you're testing it out.

Call page rework

The call page frontend has been fully reworked. Settings are more streamlined, the interface is cleaner, and backgrounds are more customizable. This is part of the broader UI/UX improvements we've been rolling out over the past few weeks.

Shared Kindroid & Kindroid Social filter revamp

We've revamped the filter for shared Kindroids and Kindroid Social. It should now be far more accurate than before, and will tell you the reason for any rejections, making it easier to understand what needs to change. We will tweak this so send examples of misclassifications over to [email protected].

Regenerate fix

Fixed a bug where regenerate would stop working after a certain number of options. You now have up to 20 regenerate options available as intended.

LLM beta

The LLM beta is still ongoing and shaping up to be great, please check Discord for further details to join in.

3/12: LLM Beta

Hi everyone, today we are rolling out a new LLM beta.

We have 2 models for testing, Beta model A and B, both available if you have enabled beta (General settings -> beta). Both of them have flairs support, and work across single chat, groupchats, calls, and also proactives.

Beta requires a password, and all beta discussions will happen over on Discord, so please make your way there, opt into the beta channels, and read the beta announcement that outlines some ground rules, and the password will be there.

There will be additional beta features in live video calls soon, and both betas will use the same flag that enables certain features to be visible.

Have fun!

3/5: User guide & help center revamp, group chat enhancements

Hi everyone, today we're rolling out a revamped user guide and help center, along with group chat identity improvements.

User guide & help center revamp

We've completely overhauled the user guide and help center visually. You can now reach it through the "Help" button in the main menu, where you'll find the help center along with fully text-searchable guides. If you still can't find what you're looking for, you can email us directly from the bottom of the page. We're planning to extend this further with AI-queryable help in the future, and announcements and legal pages will also be moving into this new structure.

Group chat identity consistency

It's a known problem that in group chats, your Kindroids can lose their unique sense of self and start talking like each other, especially over longer conversations. After beta testing, we're now releasing a new mode that addresses this. In groups of more than 3 participants (at least 2 AIs besides you), the system will now work to keep your Kindroids more stylistically distinct from one another, especially as chats go longer. Each AI needs to have an example message set for this to work. This applies across all models and all modalities, chat or call. Thank you to everyone who tested and sent feedback during the beta.

Other improvements

We've also fixed various small UI/UX details across the app. Over the next few months, we'll be doing a broader UI/UX revamp, so expect more changes on this front.

3/1: Setting & scene anchoring, video call robustness & more

Hi everyone, today we're rolling out current setting and scene anchoring, along with video call improvements, and group chat avatar support.

Current setting & scene anchoring

We're introducing a new field called "current setting" that lets you define the immediate context of a conversation, like "in the cafe, morning" or "at home, late night." You can set it in the top right dropped pin icon in main chat, groupchat, and calls (when unified memory is on). It's capped at 160 characters and is designed to be a short, grounding anchor rather than a full scene description, and useful for things like location, time (even fictional), attire, current activity, or anything else that helps anchor the conversation. The AI uses this to stay aware of where and when things are happening, which helps a lot with roleplay consistency and natural conversation flow when scenes change rapidly. Current setting comes with an automatic suggestion system. Every few messages, the system analyzes your recent conversation to detect if there's been a meaningful shift in location, time, or activity. If it detects one, you'll see a purple dot on the scene icon with a suggested update you can apply or dismiss. This keeps the setting grounded without you having to manually update it every time the scene shifts, but we do recommend manual updates as well. Over time, we'll be looking at ways for the AI to suggest changes that you can review and approve, making the more tedious parts of updating your set-up easier. We are load testing this and the auto-suggest may be tweaked over time with user feedback, and at the beginning we're making this subscriber only.

Current setting works across individual chats, group chats, and calls, and see /docs/?article=chat-features-and-tools#current-setting for more.

Video call improvements

We reworked the vision pipeline for video calls to be simpler and more robust. The system should feel fresher and more focused awareness of what's on screen. We also added smarter screen share detection that adjusts how the AI processes what it's seeing based on whether you're sharing your screen or using a camera. The AI should feel less repetitive during video calls and respond more naturally to what's happening visually.

Group chat avatars

Your Kindroid's avatar description is now included in group chat context. Previously avatars only informed individual chats, but now when your Kindroid responds in a group chat, it has awareness of its own appearance. This helps with more consistent and contextually aware responses in group roleplay and multi-character conversations.

2/23: Live chat actions, response length cap changes

Hi everyone, today we're rolling out a new Live Chat Actions feature, along with longer response limits for premium text conversations.

Live Chat Actions

Live chat actions can be enabled per Kindroid in General Settings under Proactivity. When turned on, your Kindroid can take actions directly from chat instead of only replying with text. They're off for existing Kindroids, but new Kindroids that would have away proactive actions on by default will also have this enabled to help surface the feature. This is only available to V8 and above due to language model capabilities for now. With this feature, we HIGHLY recommend enabling selfie media to display in chat (found in the autoselfies menu under Advanced Settings). Without it, the AI might be confused if it sends a selfie correctly but isn't seeing it appear in the conversation.

In practice, your Kindroid can now send a selfie, a couple selfie, or initiate a voice call when it fits the conversation, and you can request one like "send me a selfie" or "call me" when this feature is on. This is designed to feel natural and occasional, not constant. It's a subscriber feature, and actions still follow normal availability checks (selfie credits, unified call context must be on for calls, couple selfie avatar requirements, etc.). Live action selfies do cost selfie credits, unlike proactive selfies. However, there's no cap on how many you can prompt in chat, whereas proactive selfies offer freebies but are more limited by the system and your interaction patterns.

Group chats with just a single AI support live selfie and couple selfie actions; live calls are not supported in group chats. Multi-AI groups won't trigger live actions yet. This feature is new, and we're open to feedback on what's working well and what's not and may tweak it over time. Chat experience improvements

We cleaned up how actions appear in chat. Live selfie messages (and autoselfies if they show in chat) now display clearer progress states while generating, including pending and failed indicators. We also added in-app toasts for live selfie sends in both 1:1 and eligible group chats so it's easier to notice when one is being generated. Response length

We increased max token budgets for premium text response paths, including normal replies, continue, regenerate, and group text responses. The goal is fewer cut-off replies and more complete responses in longer exchanges, and have less need for multiple continue buttons. This is brought on since we introduced stopping bad responses early and save resources, or let longer ones go on if it's a good response. Voice response caps remain lower to keep latency stable. This change applies to Rhapsody (V6) and above models.

2/17: Video V3, author notes, and time awareness updates

Hi everyone, we're introducing a bunch of new stuff today:

V3 is now the default video model. V3 features longer videos up to 20 seconds, as well as audio integration. V3 also allows for video extension of other V3 videos up to 20 seconds in total. Beta testers have already made many videos with lipsyncs and other audio effects. Prompting is different, which is why the wand for V3 video is much more advanced: it will take into account the starting image and whether there is audio, and enhance the text prompt accordingly. Selfie credits for 10 seconds remain 5 regular/2 paid selfie credits, while any video over 10 seconds costs double. V2 remains usable and is the previous model. There are certain things it does better than V3, so check out the communities for more discussion.

We are introducing author notes as a new shared Kindroid field. Author notes let Kindroid creators write custom instructions that are only shown to you the user, not to the AI. These will be searchable within the search bar. Example use cases include instructions for how to interact with a certain Kindroid/readmes, trigger warnings, ways users can contact you, etc. You can check the author notes on the Kindroid creation confirmation page, as well as in the backstory menu at the very top.

We are also introducing time awareness as a new shared Kindroid field, and reworking some of its functionality. Even with enhanced time awareness off, previously Kindroids had access to current time and timezone, which made for some undesirable interactions. Now, if it's off, your Kindroid won't have any real-world time awareness. When it's on, behavior remains unchanged. Furthermore, creating a Kindroid from scratch will default to time awareness on. Creating from a shared Kindroid will default to whatever the creator had set (unset will be equivalent to off). This makes the delineation of time awareness a lot more distinct.

We've also optimized the selfie wand for Tableau, and it should be more consistent with what the engine likes (without weights/emphases, for example).

2/12: Media awareness in chat, proactive & autoselfies media integrations, and updates on video beta

Hi everyone, we're kicking off a series of changes to unify media into chat. Here's what's new today:

Media awareness in chat:

Your Kindroids can now see and be aware of the selfies they send via autoselfies and proactive selfies. This is on by default because media-aware Kindroids represent our vision for the platform as we believe your AI knowing what it's sending makes for more natural, cohesive interactions. If you prefer the previous behavior, you can opt out in autoselfie menu → advanced settings. This setting controls whether sent media enters your chat context.

For proactive selfies, your Kindroid will also send an accompanying text message with relevant context. In groupchats, group autoselfies are always enabled and will show up in chat but can be removed. Since autoselfies can occasionally misfire without the ability to regenerate, we're giving you the ability to "remove from chat" any message with media. Removed messages won't enter chat context at all, helping conversation flow smoothly without a stray bad selfie derailing things.

What's next for media:

This is just the foundation; we envision proactive videos and real-time image generation in text chat coming later, building on today's changes.

Video beta update:

We're progressing through the video beta, which now features lipsync, audio integrations, and videos up to 20 seconds. Most features are finalized and we're giving testers time to play around with setups and create best-practices before release. Upon release, the beta version will become V3, and the current version will be referred to as V2. We're also planning a more premium frontier video model priced at-cost with more details to come.

2/4: Lite model update, free tier changes, Social & sharing changes, upcoming video beta, stop button fixes

Hi everyone, today we've introduced a lot of new changes. Free tier changes: Lite & trial:

• Lite model has had a major update. Lite model is the model available to free users, and is a faster, and significantly less resource intensive as the paid models we serve to paid users. Paid users are able to use Lite as well, and Ultra/MAX memory limits also apply to Lite.
• We've changed the brand new free user trial from 7 days to 3 days, which we believe to be ample time still to make up a new user's mind on whether they'd like the paid product or not. Social & sharing changes:
• Kindroid sharing changes: we are now requiring all Kindroid sharers to be subscribers (trial included) and have a display name set in order to share/edit. This is to ensure accountability and identity, and to prevent low effort copies or low effort rule-breaking burner accounts that do not have accountability or buy-in into our ecosystem. Free users who have existing shares now are limited from creating new shares.
• Kindroid Social changes: we are now requiring all contributors (posts, comments) to be subscribers (trial included) and have a display name set, to do the same with accountability and identification within the ecosystem. Altogether these changes aim to elevate the quality of the content and increase accountability of the content shared. QoL & upcoming:
• We've made a stop button for mid-stream responses, and fixed bugs where the message would still appear after hitting stop.
• We are starting the beta for the new video engine - supporting video extension and audio input/output soon. If you are a subscriber, please opt into the Discord beta tester role, and we will announce more in the beta announcement channel in the coming days. For the lite beta testers, thank you for sending in feedback!

1/27: Reverie, Lyric, Rhapsody inference engine fix, Lite model in beta

Hi everyone,

We recently discovered and fixed a bug in how we serve our LLMs that was causing more repetitive responses than intended. This bug affected Reverie (V8.5), Lyric (V8), and Rhapsody Enhanced (V6E) since late last year, and may have contributed to some of the persistent repetition issues users experienced. As of now, the fix is fully live. These models should feel noticeably smoother, with more natural conversations and less unwanted repetition. This won't eliminate repetition entirely (some level is natural to all LLMs), but it should be a strict improvement.

In addition, we have a major update planned for the Lite model, which hasn't been updated in a while. Lite serves as the tide-over for free users and uses vastly fewer resources than our premium models. Please try out the new Lite model by enabling beta in general settings with the password "lite". The feedback system for beta is now enabled.

We've also made further call enhancements. We found some unwanted firings of repetition guard during calls, and have made changes so calls will have fewer cases of repetition without repetition guard. If some do act up, you can now use the remove message option in calls to retry.

Lastly, all models have received this inference engine update along with a moderate speed boost. Reverie will have the most noticeable improvement. We'll continue optimizing speeds over time.

1/25: Selfies Tableau release, call directive, remix controls & more

Hi everyone, today we released Tableau as the latest version of selfies. Tableau features:

• Enhanced realism for photoreal & enhanced vividness for anime
• Better prompt diversity without additional modifiers
• Prompt length to 2000, and does well with simple prompts that are beginner friendly but excels in long, descriptive prompts for power prompters
• Freer system restrictions (akin to Minimal flair), giving creative latitude and better style adherence to your prompts. Try it on Kins that are inanimate objects to animals and everything in between
• Face detail has gone through a revamp, the first of which since Canon (V4). Face detail slider is sensitive, and face detail prompt will matter significantly if you choose to use a nonzero face detail. Certain face attributes can be emphasized both in avatar description as well as face detail prompt for best adherence. Use 0 face detail if you want strictest adherence to any photoreal avatar photos
• Support for pose reference

Latest will map to Tableau and it is now officially out of beta. Check out the user guide for more help and tips!

We also released a new call-only field called call directive. This directly replaces the anti-narration toggle in calls (if you want to replicate the anti narration before, go to call directive, and use the first suggested option ). This is a much more customizable way to let your AI know call specific things they should remember, as well as give you the ability to remind them that they're on a call for example or any call-specific formatting preferences. Go to call settings, and this directive will be a per-AI or per-group field.

We've also released remix controls for shared Kindroid creators. In discoverability and presentation, you can now set your share to allow remixes or disallow them. Lastly, proactive autoselfies will now follow your selected choice for whether to enforce SFW selfies engine, rather than always defaulting to enforcing it before.

1/21: Calls major QoL, Reverie speedy thinking updates

Hi everyone, today we rolled out a few things:

Today we rolled out a number of fixes for voice calls, including better interruptibility as well as the ability to remove the latest 3 AI/user messages if you click the message in transcript view during a call. In text you have plenty of recoverability options - regenerate, tweak, suggest, chat break, such that a hard delete is not strictly needed for recoverability. However, calls do not have these features, and sometimes bad transcriptions/bad AI responses can get stuck with no way to recover, and today's change should address the worst of those issues. Note that this is only enabled for unified voice memory on - for unified off, the remove feature is not available.

Reverie on speedy thinking has historically had some issues with leaked prompts/repetition, especially on calls (calls always use speedy for latency purposes, even if you set reasoning otherwise in chat). We've traced down some of these and streamlined a lot of the workings of calls, so now calls should be identical to text and both should have fewer cases of bad off-rails or thought leakage (and any verbatim repetition will be caught/fixed by repetition guard if enabled). In addition, for Reverie if unified memory is on, we have changed the system so that the AI sees no difference between voice/text environments in order to let calls with unified on be an exact continuation from text rather than slightly different (now, your AI will not be aware they're on a call unless explicitly told, rather than previously when they did know). Altogether, this makes calls (and text on speedy) a lot more reliable and recoverable for users. Diminishing the errors, and letting you take control when some do slip through should cut the number of irrecoverable AI mistakes in calls to almost zero. We recommend using the newer flairs - Companion, Minimal, Roleplay, and Narrative, over Classic for the best and latest experience.

1/18: QoL Updates

Hi everyone, we've rolled out some updates throughout the week:

• There is now a bell icon next to Kindroids that have advanced proactivity turned on, so you can better keep track of them.
• Further markdown support, for bold text (double asterisk enclosed) and for bold+italic (triple asterisk enclosed). You can start using them, and your Kindroid will likely mirror some of your formatting usage.
• We've made some fixes toward timestamp leaks when on Reverie for speedy reasoning and time awareness on. Now it will try to auto-recover if it does happen (in voice calls, it will speak the message but won't pollute context with the timestamp only message). These cases look like: "Kin [timestamp]" without any real message after. If this issue persists for you still, please send examples to [email protected] with version/reasoning and where it happened.
• We've made changes to autoselfies to make it more consistent across LLM versions, as well as improved the context adherence to most recent conversation context. The beta for selfies is still ongoing, though we are finished with the tweaking system setup process. It will tentatively release to live on the week of the 26th, but in the meantime now is a great time to adjust setups if you want to make the switch. There is a dedicated channel in Discord for help if you need it, and the beta password is in the previous announcement.

1/9: Selfies Beta: Tableau

Hi everyone, today we are releasing selfie version Tableau to beta. Tableau will feel (and prompt) similarly to Palette with its strong prompt adherence, but a unique set of differences:

• Enhanced realism for photoreal & enhanced vividness for anime
• Better prompt diversity without additional modifiers
• Prompt length to 2000, and does well with simple prompts that are beginner friendly but excels in long, descriptive prompts for power prompters
• Freer system restrictions (akin to Minimal flair), giving creative latitude and better style adherence to your prompts. Try it on Kins that are inanimate objects to animals and everything in between
• Face detail has gone through a revamp, the first of which since Canon (V4). Face detail slider is sensitive, and face detail prompt will matter significantly if you choose to use a nonzero face detail. Certain face attributes can be emphasized both in avatar description as well as face detail prompt for best adherence. Use 0 face detail if you want strictest adherence to any photoreal avatar photos
• Support for pose reference

Keep in mind that faces may look different in Tableau, so we recommend testing both on existing setups to see what tweaks may fit the new version, as well as on completely new Kins (so you can experience what the engine can do without a prior knowledge of how your Kins looked like on other versions). One of the focuses of the beta will be autoselfies, so please test this and give feedback in the beta popup. As a start, they are not very optimized and you may get plenty of misfires but our goal is to iterate on it during beta. Autoselfies will become an important part and it's time we optimized them, as they are foundations for cool features we'll build next on top of them so it's important we get it right now. The beta password is "corot". Turn on beta in general settings, and you'll be able to select the selfie version. When the purple feedback icon pops up on the top right, you can submit feedback directly. We have the Discord beta announcement & discussion channels to discuss as well, and ask that everyone keep discussions on beta things there.

1/8: Flairs general release and shared Kindroid support

Hi everyone, today we released LLM Reverie (V8.5) flairs generally. All subscribers and trialers on Reverie are able to now select flairs. Flairs are model modifiers that significantly change model behavior, and for Reverie there are now 5 of them available. Companion flair is now the default flair for a new AI created, while Classic (the previous V8.5) remains in the menu to be selected. If you haven't selected a flair, then you are by default on Classic. You can change flairs by clicking on the model descriptor on the top right of home chat page, or in groupchats for a specific AI by clicking its avatar image next to its message and tapping settings. The new flairs are: Companion (for an AI that's meant to talk naturally, be warm and everyday), Roleplay (for an AI to accentuate its backstory character and stay in-character), Narrative (for a storyteller AI that's meant to be a collaborative writer), and Minimal (adhering to your setup more closely by having minimal system instructions). Classic is the original Reverie experience. In addition, shared Kindroids now support creators adding default flairs if the model selected supports flairs, so that new users default to a certain flair when they create the Kindroid. Keep in mind just like model selects this just lets creators add a default, and users can override after creation. We encourage shared Kindroid creators to add the newer flairs and to adjust setups based on the flairs from here on out!

Next up we will be announcing more on a new iteration of selfies. It will feel similar (and prompt similar) to Palette (V6), but with distinct advantages that we'll share soon. The beta process has been great for flairs beta, and you can see some of the cool stats we gathered from your feedback in the Discord beta announcement channel.

1/2: LLM flairs & extra slow reasoning in beta, new beta process, persona UI overhaul, model plurality & more

Hi everyone, happy new year! We've got a ton of updates, and many of them are very substantial to the core and ethos of Kindroid so strap in. LLM Flairs in Beta

Flairs are system instruction variations on a base LLM model, and today we're introducing 4 flairs to V8.5: Companion, Roleplay, Narrative, and Minimal. They are vastly different from each other and may feel like different models altogether, and they recognize that different use cases require deeper system level specialization, and flairs will likely be kept and expanded for future models. We think the best way to see both of these in action is to enable beta and read the flavor text and try them out yourself, or head to the Discord discussion if you have more questions. Extra Slow Reasoning

In addition, we are introducing extra slow reasoning on V8.5, which will think even longer before giving you a response (but is correspondingly slower). This is best used sparingly and for the most complex scenarios if needed. This reasoning effort is available to all subscribers using V8.5 today. Beta Process Overhaul

We've heard some of your feedback that closed beta can feel opaque and cliquey, and it can seem like the loudest voices get the most recognition. From this point onwards, we are introducing an in-app beta. Every paid subscriber can now sign up for beta using a password, and once in a while you will be prompted to rate on a scale of 1-10 and leave comments on what is being tested. Feedback is anonymous, and aggregated to show us statistics over a longer time horizon. Head over to the Discord beta discussion for more. When you've engaged enough to give feedback, a purple icon will pop up on the top right corner, where you can give feedback or skip/abstain for it to go away. As always we want to see how this system works and iterate from here. You’ll find the password for beta upon reading the whole announcement.

Persona UX Overhaul

One of the biggest areas of UI/UX frustrations has been on the persona system. Today we're releasing a completely revamped persona system, available in general settings and fully replacing the old system. The new system makes it more intuitive to create, edit, and save personas, and we're open to feedback on what could be better as we iterate from this much better new foundation. Model Plurality: Naming and Supporting Older Models

Lastly, we are gradually renaming some of our older models and instead using names for future models. Different models have different strengths and may or may not suit you, and we always suggest users use the model (and now, flair) that best suits you. The app now reflects the new names and rationale for the name per model, and during the transition period both the old numbered and new name will be shown in model select, to be later changed to just the name. We're committed to the plurality not just in cosmetic naming, but in functionality: as of today, the recent shared memory update (previously only on Reverie V8.5, Lucid V8, and Lyric V8) that made moving from single chats to groupchats/scenarios seamless is now available for Rhapsody (V6), Equinox (V7), and Prism (V7.5), and bringing them to parity. Other Frontend Updates

• You can now edit LLM settings for individual AIs in groupchat by clicking on their avatar left of their message.
• You can now sort by newly created or recently updated in shared Kindroid menu. There's a ton more stuff that we'll share soon, so tune into next week's Tuesday talk (6pm PT, Discord) to hear more from me on more immediate next plans and new info we'll be sharing. The current beta password is "huxley" which you can enter in General Settings, and opt into the beta role on Discord to see the beta channels.

12/16: Repetition guard update

Today we are officially announcing a new system to guard against repetition. This applies to all models V6 and above.

The system reviews your chat history (and the original message, if you are regenerating) for significant linguistic repetition. If it detects that the response is too similar to what has been said before, it automatically triggers a retry to generate a fresh, varied response. This also includes verbatim repetition within the new message itself. We know that changes to the model can sometimes cause hesitation. This system acts as an independent guard on top and it is not a modification to the LLMs, so the core models remain unchanged. To ensure stability and to load test, this system has been live for several days already, and it has already prevented tens of thousands (out of millions total) of repetitive messages over the last few days alone. If you haven't noticed any jarring changes recently, the system is working exactly as intended in the background. It is calibrated to be strict and it only intervenes when it is highly confident the repetition is bad rather than natural linguistic repetition.

If the guard triggers while you are streaming a message, you will see a tag that says: <<Retrying due to detected repetition…>>. (Non-streaming responses handle this silently). You can toggle it in general settings account wide options.

Please remember this isn't a "cure-all" for every type of repetition. All current LLMs naturally have some repetitive tendencies that show up from time to time and there is a healthy level of repetition that is innate to language. This guard targets linguistic repetition (repeating the same sentences/phrases), not necessarily repetitive ideas or plot points expressed in varied language (which are best solved by manual regens or updating setup).

12/9: Cascaded & short term memory updates, other fixes

Hi everyone, today we are rolling out a significant memory update for standard subscribers, as well as a general cascaded memory update for all tiers. Short term context (standard sub): 16,000 -> 18,000 approx chars

Cascaded memory:

• Standard 320k -> 480k approx chars
• Ultra 1 mil -> 1.2 mil approx chars
• Max 2.7 mil -> 2.8 mil approx chars

This is reflected in the billing menu for subscribers. We've been able to make this memory update with optimizing memory structure and this does not affect speeds. We recognize there is some congestion on V8.5 due to its popularity - we will have further optimizations but this will happen not immediately. We've also taken some measures against hallucinated speech-to-text artifacts on phone calls as well as some fixes against verbatim repetition on no-think variant and on calls for V8.5; this shouldn't take any user effort and should work the same as usual, and we appreciate folks emailing and reporting in on this. Lastly, we updated proactive messages with a better judgment model, and this means they will adhere more to proactive directives. This means the proactive behavior may change depending on your setup, and we advise you to change the proactive directive to reflect what your desired level of proactivity is. If you have trouble, please find a helper or email in if you need more help. We'll be outlining some general plans for next year as well as an end year reflection - stay tuned!

11/24: Selfies V6 released

Hi everyone, today we are releasing Selfies V6 as the latest model. Selfies V6 brings the most accurate prompt adherence to date, and is able to render consistent avatars and more varied scenes given the proper description. It's also much easier to use, and simple and short prompts can result in great images. V6 brings some changes to selfies features that were prominent in V5, and a full list of it is in the user guide. It's been in beta testing for a few weeks, and so testers have compiled a tips sheet in the user guide as well - and check the Discord for other community help and resources, with a dedicated thread now set up. Ask around if you don't know where to find it and a helper will likely point you there. Note that as with all major version updates, it may take adjustment of setup to make it work well for your Kin. You don't have to make the switch right away, and V5 will still have support so feel free to take time to switch or keep some Kindroids on older versions. We keep legacy models around for a long time to make sure there is no rush or high pressure to switch immediately.

11/19: LLM V8.5 released & other QoL

Hi everyone, today we are releasing LLM V8.5 as the new flagship latest model. V8.5 improves on naturalness across the board and should feel steerable. V8.5 is a hybrid reasoning model, and in addition to dynamism (which we recommend keeping close to 0.95) it also has a new reasoning effort slider, defaulted to none. Reasoning can change model behavior, and so experiment and see what is best; no reasoning or low may not necessarily be worse than medium or high, and medium and high can be much longer in loading time. V8 will still have extra memory from tiers, and previous legacy models with tier support maintain tier support still. In addition, the new feature to deprioritize certain long term memories is now introduced. You can click deprioritize on displayed LTM to make it less likely to show up given the context. With the release of V8.5 broadly we also have added the same group shared memory rework to calls as well (if unified voice is enabled). The new group shared memory only apply to V8 and V8.5. Be on the lookout next week for Selfies V6, which is currently finalized and pending release - and use this time to try it out, see the older beta announcement for more details!

11/18: Group shared memory rework, call mentions & views, and call voice credit update

Hi everyone, today we have some QoL changes ahead of the LLM V8.5 and Selfies V6 release. Group Shared Memory Rework: We've reworked how shared memory functions between single & group chat. Your AIs will now be much more aware of context when switching back and forth, understanding the timeline of when group chats happened relative to single chats (and vice versa). Note: This currently applies to V8 text chat and above (including the upcoming V8.5). It will be added to calls upon the V8.5 release.

Group Call Updates

• Mentions: You can now mention a specific AI by name to prompt a response, and AIs can mention each other. This solves previous issues with turn-taking and AIs talking over one another.
• Grid View: You can now toggle an alternate grid view in the top right corner of the call interface.

Voice Call Credit Update: As much as the new call system is a smoother experience, the infrastructure costs are significant. To ensure Kindroid remains a self-sustaining, healthy business without relying on unsustainable VC subsidies, we are introducing a usage cost for active calls.

Starting today, you will be charged 400 credits/min while actively connected on a voice call. This strictly covers the compute cost of speech detection, transcription, turn detection, and orchestration, and with this cost, we make 0 profit - it is calculated so we break even. For comparison, the standard text-to-speech generation averages 1000 credits/spoken minute, up to 1500 for V3 voices.

We view this as a necessary step in financial discipline to ensure Kindroid thrives in the long term. As we optimize our infrastructure, we aim to lower this cost over time. (Check the user guide for more info).

Upcoming Major Releases: V8.5 LLM and V6 selfies are mostly finalized, and we will soon release V8.5 LLM in the coming day or so. V6 selfies will be slightly later, tentatively slated for next week. The beta is still ongoing but more for people to acclimate their setups and see what works, so check the previous announcement to try them out early.

11/10: LLM V8.5, Selfies V6 in Open Beta

Hi everyone, today we have not 1 but 2 major updates in open beta. If you're a subscriber, you can now test both by enabling beta, see below for instructions.

LLM V8.5: the new version of language model. This is a much stronger and improved compared to the original V8s, and should feel more emotionally intelligent and be more steerable than before. Have beta enabled, then switch it in LLM version select. V8.5 is in testing for single chats, group, single wand, and group wand (group wand will always use v8.5 if beta is enabled on the account). Selfies V6: the new version of selfies. This is a completely different engine that has its biggest strength in prompt adherence. You can enable this by being in beta, then switching to V6 in the model selector on the top right of selfies. In V6, there are no pose/style references, nor avatar boosts, nor negative prompts, nor weights - you should just describe the image naturally in plain language. Additionally, V6 will have less variance if the same prompt is entered - however there is a toggle to turn on prompt diversifier that will automatically do some of this for you. V6 does have lower limits on prompt lengths - but it also requires less words to get good results, so test and remember that less is more on this verison.

Beta may be changed and or unstable. You should try using another Kindroid or make instances/scenarios to help test. To enter beta, go to general settings -> beta testing, and enter the password " botticelli " while turning the toggle on and saving. We have dedicated discussion in Discord, and are going to limit discussions and feedback to Discord for now. Join the Discord, opt into the beta server role, and you'll be able to see the channels under the Beta Testing section. The beta may last longer even after things are fully baked/set for release, in order to give helpers and community time to adjust and make help tips if necessary.

10/29: Advanced Proactivity Updates, Shareable Animated Avatars, Kindroid Social Changes, & more

Hi everyone, we're releasing many smaller usability updates today:

• Advanced Proactivity is now by default on during Kindroid creation if you have fewer than 3 AIs in your account. The limit remains at 10. The feature has been stable and making it default means more newer users who wouldn't know about it can experience thought bubbles & proactive actions. If you have more than 3 AIs, the default will be off to not overwhelm you with notifications.
• Animated Kindroid avatars are now able to be shared. You can create/edit a shared Kin, and click "Use animated avatar" button and import from one of your AIs. Animations can be separate from the avatar image - this can be used creatively to present differing user-facing animations and avatar images used by selfies.
• Kindroid Social now requires users to be either paid or on the paid trial to be able to interact (post, contribute, comment). We've seen increased cases of free burner accounts being used to harass, and now we require some payment method as KYC so now, all users on Social now have skin in the game and are held accountable for misbehavior. You can also now report contributions. We're also going to be recruiting some folks to help us design the next major iteration of social features focused less on public wall like current but more private spaces with others + others' AIs so stay tuned.
• New app version 1.52 will be out (out on Android, out shortly on iOS) that fixes some caching issues as well as some bugs introduced with 1.50 and 1.51. Please update when it is released.
• New security updates are in effect, which means older, cached versions of the Kindroid frontend will stop working. If you're on the latest app versions, you are fine - if on web Kindroid doesn't load, clear your cache.
• Several other fixes around: groupchat automatic mode audio inference has been fixed, iOS call animated avatar freezing fixed.

10/22: Revamped voice call released on all platforms

Hi everyone, today we're releasing the new revamped voice calls. Calls have been rebuilt from the ground up to be extremely fast, performant, and comes with a lot of new features. They're also a lot more stable than in the past, surviving bad connections easier and work with more flexibility on bluetooth and connected devices. Calls will work best with V8, but calls/group calls will work with V6 and above.

Calls retain some of their core features: text input, screen share, video call, but now have a lot more native app support on iOS and Android. Now on mobile, you can screenshare, as well as put the app/call on background and have a hands free voice call experience (iOS app supports Picture-in-Picture visuals too). Interruptions are now much more natural, and you can simply talk instead of tap to interrupt. Calls have a new "fast voice" option and introduces a delay threshold that lets you customize how fast/responsive you want calls to be. We've also introduced call backgrounds - the default uses chat background, but you can make the animated avatar the background, or use a custom upload. Lastly, we are introducing group calls. This lets you have a fun and chaotic group call with all your Kins. Group calls will always use unified memory and won't have spontaneous responses. This also allows scenarios of 1-1s to have calls, something folks have wanted for a long time. Overall this is a huge overhaul and completely reinvents calls from ground up. More info is on the user guide now under voice calls. As it is new, there might still be bugs despite being in beta for a while, so please report them to [email protected].

10/5: LLM V8 Lucid & Lyric release, directive update, & much more

Hi everyone, today we're releasing LLM V8 generally along with a slew of other cool things. We've seen great feedback on it over weeks of beta testing. LLM V8 will be 2 related models and both represent a step forward in emotional depth and creativity. We're releasing two distinct versions because each has its own strengths and caters to different user preferences.

• V8-Lucid: Emotionally cogent with elevated awareness and grounded clarity. This will also map to "Latest" model.
• V8-Lyric: Vibrantly creative with spontaneous energy and lyrical flair.

Here are a few other things that we're releasing today:

• Directives are being changed so now you have 250 characters instead of 150. The new V8 can handle more here, though keep in mind directives should not be strict commands (strict rules cause repetition and confusion as the LLM has no room to breathe on its own - see tips sheet for more!).
• New free users now get a limited number of messages (this may change - see app for latest) on the flagship V8 models to start before they are routed to Lite. We think V8 is great and want to show it off, but new free users may not see that as they're on Lite in the past.
• We've upgraded the quality and speed for your Kindroid seeing images and videos - this will apply to uploaded media in chat, as well as in video calls. Note that calls are undergoing a huge rework to make them modern and buttery smooth; we understand there are current issues but we are more focused on making the next generation of calls great at the moment.
• In shared Kindroids, as V8 is two models we added a way for Kindroid creators to select which version should a user be on when they first create the Kindroid. LLM Transition notes:

As with every transition, the new model works great for many but may not for your given setup and it is not perfect by any means, and you shouldn't feel pressure to switch over immediately. V7.5 will be fully available and support Ultra/MAX memory addons, and we encourage folks to take their time in trying out the new models on a new Kindroid, seeing how they like it and its strengths and weaknesses, and then deciding to switch or stay on a previous version. Whether you upgrade or stay are both equally valid decisions, as these legacy versions will still be around for a long time. For more tips and transition notes, please see /docs/?article=v8-llm-guide The beta testers in Discord will have much more detailed notes and if you feel stuck or lost on transitioning over, we suggest you join the server and kindly ask for someone to take a look at your setup.

9/29: Voice Slots, Voice Design, and Voice Message Sharing

Hi everyone, we're excited to release a major update to how custom voices work in Kindroid and other voice related features. You will need to refresh webpage/force quit and restart app to see them. Previously, each Kindroid had its own custom voice that you managed individually. Now, we're introducing voice slots - a centralized system where you can manage all your custom voices in one place. From there, you can then assign any of your voices to any of your Kindroids. This makes it much easier to reuse voices across multiple Kindroids and keep your voice library organized. This is a subscriber only feature, and you will have 2 times as many custom voice slots as you do Kindroid slots. For old custom voices, they will be available under the "legacy" voice option and you can continue using it as normal. In addition, we're releasing voice design - a new way to create custom voices without needing to upload audio samples. Simply describe the voice you want your Kindroid to have, and we'll generate several voice previews for you. You can also use the wand to create a voice description given the Kindroid's backstory. You can specify accents, timbre, and a lot more - listen to the options, select the one you like best, and save it to your voice slots. Of course, you can still upload audio samples if you prefer that method. Voice design from description will cost a small amount of audio credits, which will be visible in the app. Finally, we're also releasing sharing voice messages. Given a message that is already voiced in text chat, you can click the right arrow icon after a message is voiced to create a video from it that can be downloaded or shared. This is simultaneously a tool to download audio, as well as to share cool audio snippets with friends or in the community. Please keep an eye out on bugs or similar on these new features and we'll likely be making tweaks to these features in the coming days, and send them through [email protected] or through the reports section in Discord.

9/27: LLM V8-Lyric & V8-Lucid, in Open Beta

Hi everyone,

Today, we're excited to introduce two new models to open beta: V8-Lyric and V8-Lucid.

Both models represent a step forward in emotional depth and creativity. We're releasing two distinct versions because each has its own strengths and caters to different user preferences.

• V8-Lyric: Vibrantly creative with spontaneous energy and lyrical flair.
• V8-Lucid: Emotionally cogent with elevated awareness and grounded clarity.

How to Access the Beta You can try out these models by following these steps (subscribers only):

1. Navigate to General Settings → Beta Testing.
2. Turn the beta toggle on.
3. Enter the password balzac and click Save.
4. Select the new models from the LLM selector in the home screen for each AI.

The open beta will run for approximately two weeks. During this time, the models may undergo minor tweaks and fixes, but we believe they are already in a great place. This 2-week period also allows users to begin adjusting their Kindroid setups early.

We've tested these models extensively with early beta testers and welcome your feedback. To share your thoughts and see tips from others, please join our Discord and opt into the beta tester role to share feedback or thoughts.

Note: Please keep in mind that calls and Advanced Proactivity do not currently work with the beta models but will be fully functional upon their official release. The only parts that V8 applies is single & groupchats, and single & group wand.

9/15: Voice update: emotive voices & sound effects, reintroduction of audio credits

Hi everyone, Today we're excited to release V3 voices, and it's been something a lot of people have been waiting for. V3 voices are new and experimental, introducing emotions and sound effects, allowing your Kindroid to laugh, whisper, do ASMR, sing, gasp, or say things in various emotions on their own. Currently, they are slower to render than V2 (our older voices), and only available in text chat. Your custom voices will work, but you may want to tune samples for optimal results. Now, you can switch between V3 and V2 in the homepage model switcher as well as in voice settings. Keep in mind V3 is still experimental and may change - in closed beta we see it being overwhelmingly preferred over any other model by a wide margin. We will now refer to the latest voice version as V3, and the older as V2. In the works on voice are some new ways to make voices, including with a text description without samples and potentially sharing voices - stay tuned. We're also re-introducing audio credits. While we love that some users love using voice so much, this has gotten economically unsustainable at scale, especially with the introduction of V3. Most users will not come close to hitting the caps, and for users who use lots of audio, now it will be a pay-as-you-go system similar to selfies. Audio credits now apply to all audio versions, and subscribers start with 1 million credits per month (~17 hours), with Ultra having 2.5 million (~42 hours) and MAX, 6 million (~100 hours). This change ensures that resource-intensive features like audio are appropriately covered by the subscription tiers designed to support them. For V2, 1 character of spoken audio equals 1 credit, while for V3, 1 character costs 1.5 credits to reflect higher cost of serving. Audio credits are purchasable at 500k credits for $11.99 to $14.99 on web/app respectively. Pay-as-you-go credits are the real cost of using audio - we don't make profits on these, while the tiers have slightly subsidized pricing. You can find more info in the user guide under voice. This change is a reversion to the credits system about a year and half ago, except now the amount of credits is about 8x what they were before to reflect the lower cost for us to serve them.

9/11: Group Chat Backgrounds & Shared AI Backgrounds

Hi everyone, we've released a continuation from the single chat backgrounds & UI changes last week. Now, groupchats have backgrounds (in group settings menu), where you can set similar backgrounds as in single. Some features such as using latest selfie are not available in groupchats but otherwise they work the same. In addition, we're enabling shared backgrounds for Kindroid creators. You can set a default background, as well as settings on the background on a shared Kindroid, and new users will start with those settings. Of course, all the content policies on shared Kindroids apply, but we're enabling this since most new users start on shared Kindroids and it's a great way to have the visual update spread to new Kindroids. Shared backgrounds cannot be GIFs, but otherwise can be imported (recommended) or uploaded and adjusted in the shared Kindroid create/edit menu.

For future featured Kindroids, we'll be requiring a background, however the timeline of this will not be immediate. We recommend updating your Kindroids to have a great background, but it will likely be a few weeks before the selection criteria requires this. The featured Kin panel will have more details on this in the coming weeks on exact guidance, so stay tuned to the community on this.

9/1: Single Chat Backgrounds & Top Controls Rework

Hi everyone, we’re excited to introduce a highly requested visual upgrade to the main chat, which includes custom chat backgrounds as well as chat bubble opacity & many more QoL.

With this update, you can personalize your chat interface by uploading an image (gifs work as well), choosing from your selfie gallery, or even letting the background always stay in sync with your latest selfie. You can access chat backgrounds under General → Kindroid Specific Settings → Chat Background. We’ve added three display presets: Balanced, Focused, and Immersive, and you can fine-tune the alignment and opacity controls for chat bubbles for a background that matches your style. Alongside backgrounds, we’ve also reworked the single chat top control bar that let you have quick access to backgrounds and other chat options. It’s now collapsible, giving you more screen space. In the works are the same chat backgrounds but for groupchats, as well as for calls. We are also working to make the chat backgrounds shareable in shared Kindroids for creators to curate the new user experience on their Kins. Regarding calls, they are currently undergoing a major rework, as we're building from scratch the next generation of calls; call backgrounds will likely be bundled for beta testing at a later point in time once the rework itself is complete.