A viewer's guide / how to play along

MOONLASH

Moonlash is an advanced interactive 3D VTuber

WHAT CAN
MOON DO?
01 / you act
You send a messageChat, join voice, post an image, trigger a redeem, or start a streak.
02 / she thinks
She remembers who you areShe pulls up your personal history, previous conversations, and active scenario context.
03 / she plans
The queue schedules your eventThe system prepares assets in the background so there's no lag when it's your turn.
04 / she acts
The scene comes aliveShe speaks, moves, reacts to your images, plays your song, or starts a game.
05 / she remembers
The memory is savedShe writes your turn into her permanent history and can bring it up days later.

Interactive features

How the audience controls the show.

Category key
intelligence live input presentation continuity experiences
Conversational memory 01

She remembers your past conversations.

She keeps a running memory of who you are, what you like, and your shared history. Mention something you told her last week, and she'll reference it naturally during conversation.

your nickname
your comments
your personality
past stream logs
Lore & Personality

Deep Personality.

Moonlash has a deep and rich personality and backstory lore, as well as organic experiences shaped by her interactions with the audience.

Queue

No message is dropped.

Your chat, redeems, and voice comments enter a fair queue so everyone gets their turn, even during high-traffic moments.

Preparation

Zero-lag media loading.

When you trigger an image or song, she prepares it in the background so the stream doesn't freeze while waiting.

Coordination

Speech and movement match.

Her facial expressions, hand gestures, body posture, and camera angles align dynamically with whatever she is saying.

Recall

Search her brain.

Use commands to ask her what she remembers about a topic, and she'll search her entire stream history for the answer.

Agents

Moon uses agents.

Since Moon has so much going on, so many subprocesses to manage, many of them make use of subagents to solve problems.

Focus

She stays on track.

She automatically cleans out old background spam and chat noise so she can focus on the current topic you are discussing.

Safety

She doesn't lose your turn.

If the stream reconnects or crashes, your pending redeems and her conversational thread are restored from her latest backup.

Continuity

The history is real.

Every conversation you have is logged, creating a continuous historical record that spans months of live streams.

Multimodal input

She experiences your world.

She listens to your voice, reads your chat, checks your images, watches your video clips, and reacts to in-game events in real time.

Voice chat

She hears you clearly.

She knows when you start talking, waits for you to finish, filters out background noise, and responds directly to your voice.

Vision

Show her your images.

Post an image, and she'll inspect the visual content, analyze what's happening, and give you her thoughts aloud.

Voice selection

Many natural voice styles.

She can swap speech providers if one lags, giving you uninterrupted, highly expressive voice responses.

Audio

Clear sound separation.

Her voice, the music, your voice, and sound effects play on separate channels so nothing gets clobbered.

Playback control

Pause, skip, and stop.

If her response goes on too long or you need to interrupt, she can pause, skip, or stop speaking immediately.

3D world

A living, persistent physical space.

She lives in a complete 3D environment with a map, lights, physical props, clothing options, particle effects, and collision boxes. Your actions physically alter this world during the show.

Expressions

Accurate emotion mapping.

Her facial expressions and poses change based on the mood of her conversation, shifting naturally as she talks.

Camera

Cinematic angles.

Special events, games, and performances can take control of the camera, zoom in, frame the action, and restore the default view afterward.

Focus locks

Unbroken performances.

When she's dancing, singing, or playing a game, the system protects her timing so other actions don't interrupt the scene.

Connection

She reconnects automatically.

If her 3D world lags or restarts, her connection clears pending junk, resets her face, and resumes within seconds.

Lag protection

Butter-smooth frame updates.

The system monitors data flow and discards outdated movements so she never lags behind her voice.

Inspection

We see what she sees.

The operator can check her current 3D position, rotation, camera angles, and animations directly from her system logs.

Expressions

Timed facial expressions.

Her emotional reactions are queued, played, held for a natural duration, and cleanly reset to a neutral look.

Animations

Ordered body movements.

Her physical actions play in sequence, avoid clashing, and transition smoothly back into her idle movements.

Routing

Control other characters.

Your actions can animate secondary characters, guests, or mascots in the scene without clobbering her movement.

Missing assets

Tracking what's missing.

If she requests a pose or expression that isn't in her assets, the system logs it so the crew can add it later.

Props and effects

Objects react to her actions.

Guitar jams, katana slashes, confetti, sleep modes, and animation speed changes sync automatically with her physical actions.

Caller pictures

See who's on the line.

When an AI calls her, a custom profile picture of the caller can be generated, displayed on screen, and cleaned up afterward.

Avatar state

Sleep and speed control.

She can go to sleep, wake up, or speed up her movements based on the pacing and energy of the conversation.

Food and drinks

She eats and drinks props.

Give her a drink or a snack, and she'll hold the physical object, execute an eating animation, and react to the flavor.

Channel-point rewards

Turn channel points into stream segments.

Triggering a reward prepares the voice, generated images, overlays, camera moves, and sound effects, running a staged, unbroken show segment.

Fast actions

Tactile, immediate reactions.

Give her headpats, trigger an explosion, throw a meteor, change her outfit, or make her spin around with zero delay.

Generated media

Generate and inspect art.

Ask her to create an image, and she'll generate the art, inspect the result, display it on screen, and discuss it with you.

Interactions

Target different characters.

You can target Moonlash, the streamer, a guest, or the mascot with physical and verbal interactions that produce unique responses.

Costumes

Swap her outfits.

Change her visual outfit into a bunny, cat, or demon form. The system updates her prompts so she remembers her active costume.

Conversations

Ask her anything.

Request her hot takes, opinions, ratings, or roasts. Ask her to highlight your message, and she'll address you directly.

Soundboard

Live stream sound cues.

Trigger applause, impacts, stingers, music shifts, and rejection sounds that play in perfect sync with her physical movements.

Performances

Dance and song sequences.

Coordinate music tracks, dancing animation, camera movement, and speed shifts, returning smoothly to normal state when over.

Media reports

Staged review formats.

She can deliver a stylized movie review, product review, news report, advertisement, or vertical phone wallpaper.

Utilities

Greetings and status checks.

Check your remaining image quota, say your daily hi, or check timezone clocks using conversational commands.

Modular registry

Seventy plus custom scripts.

Every interaction is an isolated, independent script. If one script has a bug or fails to load, the other seventy continue working perfectly.

Execution

Delayed execution staging.

Expensive image generation, speech, and model work prepare silently in the background so you never experience frozen frames.

Timing

Fast actions get priority.

Tiny visual effects execute immediately, while longer cinematic sequences wait safely in queue behind active performances.

Restart

Restoring pending actions.

If the stream needs a quick reboot, your pending queue items, username details, and prepared images survive the transition.

Output

Publishing on many platforms.

Her actions update her 3D world, local speakers, chat, overlays, leaderboards, community channels, and stream logs simultaneously.

Access

Fair usage and restrictions.

The system enforces user cooldowns, image quotas, and active-game boundaries so no single viewer can monopolize her attention.

Collections

Roll for randomized waifus.

Draw randomized characters and products. She'll generate their portrait, read their background, voice their lines, and give her thoughts.

Audience games

Play Family Feud with chat.

She'll generate questions and boards. Type your guesses in chat; she'll score them using smart synonym-matching, play applause, and crown the winner.

Media

Timed media playback.

Music, clips, effects, generated speech, applause, and transitions are coordinated with the visual scene.

Publishing

Audience-facing result output.

Images, outcomes, messages, timers, and announcements can be shown to the audience in several places.

Staged productions

Full broadcast review segments.

Trigger a full segment (movie review, news report, or ad), and she'll generate the content, inspect it, present it, update her overlay, and restore her camera.

Queue pacing

Smart presentation timing.

Staged media stays visible for a minimum duration but can shorten automatically if the queue is backed up with other work.

Cleanup

Automatic scene restoration.

All temporary files, images, audio streams, overlay changes, camera shifts, props, and timing locks clear out when an interaction ends.

Model laboratory

Testing models against real stream history.

Before any model gets to speak on stream, we test it against a real saved memory slice. We compare dozens of providers and endpoints for latency, personality, and accuracy.

quality
speed
style
cost
Method

Testing under real stress.

We run multiple parallel tests on every candidate, measuring wall-clock timing, capturing text responses, and averaging results.

Judgment

Latency is only one score.

We test for response speed alongside natural voice tone, visual control tags, duplicate avoidance, and consistent character personality.

Reliability

Self-healing runtime.

The system recovers automatically from broken internet connections, speech failures, microphone glitches, and community disconnects.

Guardrails

Automatic output correction.

The system monitors her responses, detects when she leaks prompts, repeats herself, ignores tags, or speaks gibberish, and silently triggers a rewrite.

Monitoring

Watching her health.

The operator monitors load times, API costs, queue length, missing animations, repetition statistics, and active memory sessions in real time.

Coordinated runtime

The core performance engine.

She coordinates her speech, facial tracking, poses, cameras, lighting, props, particles, and secondary characters so the stream remains one continuous show.

Secondary characters

Spawn other characters.

Your actions can spawn guests or royal guards who load their own personalities, voices, and animations into the active scene.

Operator control

Tools behind the curtain.

The crew monitors live subtitles, ad timers, custom overlays, announcements, direct camera switches, and system status logs.

Phone calls

Call characters on the phone.

Call her or other characters. The phone rings, she answers, she reacts to your voice, she displays your picture, and she wraps up before hanging up.

Phone logic

Call state and rules.

Active calls lock the camera, prevent other calls from clobbering the line, route the conversation into context, and hang up cleanly.

Chat mascot

A dedicated stream mascot.

Our sassy, pink-haired mascot speaks with her own voice, moves, and reacts directly to chat while Moonlash remains the primary host.

Ad breaks

Staged commercial breaks.

Ad breaks coordinate countdown timers on screen, start and end sound effects, background music, queue pauses, and stream returns.

Hotkeys

Direct keyboard triggers.

The operator can trigger physical props, camera zooms, scene changes, sound cues, or panic stops directly from their keyboard.

Socket transport

Responsive overlay messaging.

Overlays and scene layers exchange light events for timers, subtitles, game boards, song lists, and system status.

Local API

Local hardware controls.

External tools, hardware keys, and Stream Decks control her 3D world, queue order, camera presets, and media directly.

Live platform protocol

In-house connection client.

Our custom connection handles platform logins, subscription validation, automatic reconnects, chat messages, follows, cheers, raids, and channel points.

Discord presence

Always online, 24/7.

Even when the stream is over, she lives in Discord. You can discuss things with her, ask for custom images, play games, or chat directly in voice channels.

Discord linking

She knows who you are even on Discord.

Moonlash uses a system where Khaos links your Discord account to your Twitch account name profile, allowing her to recognize you and maintain context across platforms.

Discord text

Channels, DMs, and mentions.

She monitors specific channels, responds to direct pings, answers private messages, and reads references to earlier chat lines.

Discord context

She reads the conversation.

She reads message history, checks referenced lines, maps user nicknames, and resolves user profiles to keep track of her friends.

Discord images

Post pictures, GIFs, or videos.

Attach media, and she'll inspect it, summarize what she sees, and include your upload as context during discussion.

Discord embeds

Review shared video clips.

Post a shared clip or thumbnail, and she'll check the title, author, description, and preview image to chat about it.

Discord art

Request custom images.

Use Discord commands to ask her for a specific style. The system tracks your quota, updates her memory, and posts the final art.

Discord commands

Control her from Discord.

Commands let you look up your remaining quota, timezone clocks, active character biographies, or search stream history.

Discord voice

Talk to her in voice channels.

Join her voice channel, speak aloud, and she'll listen to your voice, identify who's talking, wait for you to finish, write your words into context, and play her voice back in real time.

Voice input

Clean speaker recording.

She records each speaker's turn separately, filtering out empty silence, spotty network lag, and bad audio clips.

Voice output

Converted high-quality speech.

Her vocal synthesis converted cleanly into the voice channel ensures her responses sound clear and natural.

Characters

Mouth controls on screen.

Mention a specific screen character, and she'll route the speech to them. The character responds with their own voice and synchronized lip sync.

Stream modes

Stream, watch-along, and offline.

She knows when the main stream is live, when she is broadcasting from Discord, or when she is ambiently available to keep things coordinated.

Integrated runtime

Asynchronous subsystem coordination.

Chat, voice, vision, games, timers, controls, model calls, media preparation, character state, and scene commands are scheduled through one runtime with explicit priority and interruption rules.

Conversational intelligence

Persona, memory, retrieval, and response control.

The conversational layer maintains long-running context, user memory, scenario state, model routing, validation, repair, summarization, and structured history.

Multimodal perception

Voice, sound, image, animation, video, and game input.

Several audio and visual pipelines convert live activity into normalized context, with turn detection, filtering, frame analysis, game-specific prompts, and low-latency paths.

Tracking

Mapping Animations

ARKit 52 blendshape facial tracking, 15 Visemes blendshape lipsync, detailed finger precise hand tracking, body armature enabling tracking and humanoid animations.

3D production environment

Map, characters, cameras, lights, and physics.

Independently controlled characters, clothing physics, props, particles effects, collisions, ragdolls, lighting, and cameras.

Creative prompting

Moon comes up with prompts

Many features allow Moon directly to write image prompt herself and then to show off the result of those generated images.

Character systems

Host, streamer, screen characters, phone characters, and mascots.

Each character type has separate context, voice, lip sync, animation, expression, tracking, memory, scene commands, and audience interaction behavior where required.

Audience systems

Immediate reactions, staged interactions, games, and persistence.

Audience actions range from short physical or audio effects to generated-media productions, randomized collection systems, phone calls, timed games, performances, and cinematic attacks.

Media production

Speech, music, visual generation, overlays, and publishing.

Generated speech, images, animations, videos, full songs, reports, reviews, subtitles, timers, and results are prepared, presented, routed, published, and cleaned up.

Operations and recovery

Hotkeys, control APIs, sockets, analytics, and restart state.

Operator controls, platform events, local endpoints, overlays, reconnects, provider failover, queue recovery, backups, telemetry, and watchdogs support continuous live operation.