October 2026
Learn LoreWeaver in 21 Short Videos
- A new video course: 21 short lessons take you from your first reply to publishing a world on the Hub, each showing one feature on screen in about a minute
- Every lesson was recorded on a computer and on a phone, so the buttons are right where you’ll find them. Numbered steps beside each video let you jump to the part you need, and the transcript follows along as it plays. Open Tutorials from your account menu, the Hub or the feature guide
- Import whole chats from Character.AI, SillyTavern, Oobabooga and FictionLab, with every message in order. Each chat becomes its own story
- Come back to a story after a few days away and a short “Previously” card reminds you where you left off
- When you update a published world, you can choose not to show players what changed. They see that an update is ready and download it when they like
- If a world can’t be saved because something in it breaks a content rule, the message now tells you which part and which rule. Lore cards are checked the same way when you save them
- The import window opens on the apps it reads, with every format under Details
- The create box stays above the keyboard on phones
- If signing up from a guest story fails, you now see why instead of nothing happening
- Updated Terms of Service
- Safety and security updates
Your Story, Remembered in Its Own Words
- Your storyteller can now look back at the story itself as well as its memories. When a promise or a tiny detail from hundreds of replies ago matters again, it finds the scene where it happened and reads it as it was written
- It searches by meaning and by the exact words you use, names included. In our tests on long stories, it answered up to three times as many questions about earlier events correctly
- Stories you already have get it too, and their earlier chapters fill in as you keep playing
- The story recap now refreshes every 16 replies instead of every 4, since the storyteller can read the scenes themselves. Remember now in the Memory panel still updates it right away
- Editing, deleting, rewinding or branching updates what the storyteller looks back on
- The story recap no longer skips replies in very long stories
- Memories made at the same moment stay in story order, in branches too
- Editing a message on a phone keeps the box in view above the keyboard
- Memory updates no longer stall when lots of them happen at once
- Regenerating or switching versions of a reply no longer hangs in long stories, and neither does branching
- Safety and security updates
Select Many & Edit in Place
- Select as many Loom threads or lore cards as you like, then turn them on or off, move them, export them or delete them in one go. Lore cards can be pinned together too
- On a phone, press and hold a Loom thread to start selecting
- Editing your message or the storyteller’s happens right in the story, in a box that grows as you write, and opens full screen with one tap on any device
- In Manage publication, import your lore export to update a published world’s lore. Cards are matched by name, so changed cards update and new ones join without doubles, and you see every change before you save
- A published report on how LoreWeaver’s story memory does on LoCoMo, a public benchmark for long conversations
- The push notification switch now sits in Account, beside Haptics, in the iOS and Android apps
- Back in the world builder takes you to your idea with everything you typed, including your phone’s back button
- Push notifications on iPhone no longer get stuck off
- Comeback emails no longer say your own character misses you
- Dropdowns look like the rest of LoreWeaver everywhere, including Windows
- Loom’s Trigger scene image action is gone, since it never made a picture
- Safety and security updates
The Loom, Rebuilt & a Studio for Publishing
- Loom threads can now do math and compare any two values, and short scripts handle bigger ideas, like a shop that checks your gold before it sells
- Test a turn to see exactly what every thread would do, without touching your session
- A rebuilt Loom editor: each thread has its own page, and the Variables panel renames a variable across every thread at once. It all works on phones too
- Ten new Loom templates, 53 in all, each with its own icon. Every thread you have already built keeps working as it did
- Review & publish is now a studio, with a live preview of your Hub card and a checklist
- Your world’s page is a showcase you build from blocks, from lore cards to galleries and video, started from your world’s own characters and places in one click
- The new model guide shows how each model handles roleplay, with up to four side by side
- DeepSeek V4.1 Flash now runs DeepSeek’s October update, with stronger replies that start in about a second
- DeepSeek V4.1 Flash is the new Best overall pick on Ultra and Legendary
- In worlds set to short replies, DeepSeek V4.1 Flash keeps to the length and ends on a full sentence
- Memories and the story recap hold on to more of what happened in your story
- Suggested moves and lore card descriptions fit your story better and arrive faster
- Auto scene illustrations pick their moments better and follow the scene more closely
- Import up to 50 PNG, JSON or CHARX character cards at once from My Stuff or the import page, each as its own character world
- Imported cards keep their formatting and emoji, and SillyTavern’s Character’s Note comes along too
- The write-it-yourself form uses the builder’s cards and its Tone and Writing style choices
- Loom’s “Input mode is” and “Lore card is inactive” conditions now work, and scheduled threads wait for their turn
- On Continue, Loom threads no longer mistake the storyteller’s words for yours
- Safety and security updates
Step Outside the Story
- A new OOC mode in the composer lets you step out of the story and talk to the storyteller directly. Nothing in the story moves, and Loom, memories and the recap leave the exchange out
- In your own worlds, the Context Inspector shows the names of threads you have hidden. Players and playtesters still see them hidden
- Space Bunny Alpha’s preview has ended. If it was your pick, you are now on DeepSeek V4 Flash
- Notes typed as “(OOC: …)” in Story mode count as part of the story again, so a director’s note like a timeskip moves the story and Loom keeps running
- The account menu scrolls on short screens instead of cutting off Edit Display Name and Sign out
- Safety and security updates
Gifts & Fairer Plan Changes
- Give anyone 1, 3 or 12 months of any plan, paid once. Nothing renews, they never need a card, and the code never expires
- Upgrading on the website charges only the difference for the rest of your billing period, and moving to a smaller plan waits for your renewal
- Cancel from Usage & Billing in a couple of taps, and turn renewal back on any time before your plan ends
- Gemini 4 Argon, Google’s newest flagship storyteller, on Ultra and Legendary for a limited time
- Worlds with lots of Loom threads hold on to your latest scenes, even alongside a big lore book
- The Context Inspector shows which Loom threads shaped a reply
- DeepSeek V4 Flash keeps better track of earlier events in your story
- Guests can save their story with Google, Apple, Discord or email right where their free replies run out
- Updated Terms of Service, and our legal pages now lead with the LoreWeaver name
- Loom threads stay in the order you added or arranged them
- Tabs left open for days no longer crash after an update or reload over a reply you are typing
- Tapping Create twice in the world builder no longer makes the world twice
- The plans page shows the real memory limits for every plan
- Safety and security updates
A Brand-New World Builder
- A brand-new world builder: describe a world or a character and watch it come together in front of you, with its places, its people, a painted cover and an opening scene
- Change anything by editing it in place, or just tell the builder what you want different. It builds exactly what you ask for, with no presets
- Prefer to do it all yourself? Write your world from scratch instead
- Every Hub world shows how many messages players have written in it
- The Everyone, Teen and Mature setting now lives in Content Preferences instead of on the Hub and Search
- When a published world’s rating changes, its author gets a note explaining why
- Muting a voice call mid-sentence now sends what you already said, and the popping some players heard on calls is gone
- Updated Terms of Service and Privacy Policy
- Usage & Billing and your subscription are back near the top of the account menu
- Lore cards no longer go stale or drop out mid-story, the Loom fires when it should, and out-of-character messages are handled properly
- Loom threads woven into a reply wrap instead of running off the screen
- The story no longer shakes while you type a long message
- Some models no longer leave their thinking in your replies
- Saving changes to a published world no longer fails when its lore book is huge
- New & Noteworthy lists the newest worlds in the order they were published
- Safety and security updates
Speed Upgrades for Every Model & Better Voice Calls
- Faster Continue and Regenerate: on every model and every plan, both start writing noticeably sooner
- Several storytellers reply faster and fail less often
- Voice calls respond as soon as you finish talking, and your character starts speaking while the reply is still being written
- The call voice flows naturally from one sentence to the next, and short stage directions in italics are no longer read aloud
- Calls show the world’s full cover image, and scene images made during a call appear as pictures in the captions
- Guest play opens right on a world’s opening scene with a premium storyteller, and guests get more replies before they need an account
- A refreshed look across the website and app
- Verification code emails now come from LoreWeaver in our own design
- Grok 4.7 replies no longer fail
- DeepSeek V3.2 no longer cuts off replies in some scenes
- Gemini 3.7 Flash replies have a backup, so a busy moment no longer ends in an error
- On desktop, the lore category bar has arrows so you can reach every category
- Security updates
Double Context Weekend & a Leaner Model Lineup
- Double Context Weekend: Gemini 3.8 Flash and Gemini 3.7 Flash reach 32K on Pro, 96K on Ultra, and 160K on Legendary, and DeepSeek V4 Flash runs from 32K on Plus up to 200K on Legendary. Look for the 2× in the model picker; Deep Context rewards stack on top
- Faster replies: storytellers respond slightly faster across most models
- Qwen 3.8 Max is improved and more reliable
- Five little-used storytellers retired, with stories switching over automatically: Gemini 3.6 Flash and Gemini 3.5 Flash Lite to Gemini 3.8 Flash, Qwen 3.7 Max to Qwen 3.8 Max, Aion-2.0 to Aion-3.0 Mini, and Seed 2.1 Turbo to DeepSeek V4.1 Flash
- Prism Ultra no longer fails replies when one of its storytellers is busy
Custom Lore Categories, Card Import & Play Without an Account
- Custom lore categories: create categories like Spells or Noble Houses, and rename, reorder, or remove any of them, the built-in six included. Removing one asks where its cards should go, and Lorekeeper files its suggestions into your categories
- Character card import: bring characters from SillyTavern, Chub, and Janitor AI as PNG, JSON, or CHARX. Greetings become openings, lorebooks become lore cards, and the Lore tab imports a card into a world you already have
- Swipe to regenerate: swipe left on the latest reply for a fresh take and right to go back, or use the arrow keys on desktop
- Play without an account: anyone can press Play now on a published world, and their story comes with them if they join
- No more hard stops: reaching your daily premium limit continues the scene with the free storyteller, with a heads-up before your last premium replies, and free players can rewrite one reply a day with a premium storyteller
- Usage page: Usage & Billing in the profile menu shows your allowances, reset times, and a calendar of your year of play
- A cleaner story screen with refreshed dialogs and menus, and a session list sorted by when you last played, with branches tucked under their story
- Swipes and regenerations answer your last message again instead of jumping the story ahead
- Rewind, undo, and clear roll back Loom’s changes too, and branches keep the Loom state from where they split
- Loom threads imported from another world now point at your own lore cards
- If a renewal payment fails, your plan stays active for a week while you sort it out
September 2026
Aion 3.5 & Space Bunny Alpha
- New model Aion 3.5 (Ultra+): a roleplay-focused storyteller for layered motives, dramatic conflict, and sustained tension. Up to 16K on Ultra and 24K on Legendary
- New model Aion 3.5 Mini (Pro+): a lighter roleplay specialist for character-driven scenes and rising stakes, now Recommended in place of Aion 3.0 Mini. Up to 16K on Pro, 32K on Ultra, and 48K on Legendary
- New model Space Bunny Alpha (Plus+): an experimental voice with up to 1 million tokens of context while in Alpha. Your plan’s usual message allowances still apply
- Space Bunny is an anonymous-provider preview: prompts and replies may be retained by its provider, but are not used for training
Your Worlds Have a Voice
- Character calls rebuilt with a cleaner design, a spacious desktop layout, and a more polished mobile experience
- Speak naturally, interrupt a spoken reply, and keep the conversation moving with more responsive turn-taking
- Live sound waves react to your voice and to spoken replies, with clear listening, thinking, and speaking states
- Captions are on by default, with clearly labeled speakers in order
- Failed spoken replies can be retried, with clearer guidance when your microphone disconnects or pauses
- A faster voice engine powers calls and Read Aloud, and audio starts playing while the rest is still being prepared
- Pause, resume, or stop narration right beneath a story reply
- Voice storytelling now works in regular worlds too: speak your actions and dialogue and hear the narrator continue, with the conversation saved in the same story
- New character and story worlds start with a voice selected and Read Aloud on; existing world preferences are unchanged
- Additional security improvements protect promotional rewards and strengthen abuse prevention
Four New Models & a Lighter LoreWeaver
- New model GPT-6 Luna (Plus+): clear prose, focused scenes, and carefully directed dialogue. Replaces the previous Luna and now starts on Plus
- New model MiMo-V2.6 Flash (Plus+): a nimble choice for unfolding adventures, quick exchanges, and busy scenes
- GPT-6 Luna and MiMo-V2.6 Flash offer up to 16K on Plus, 32K on Pro, 64K on Ultra, and 100K on Legendary
- New model MiMo-V2.6 Pro (Pro+): interwoven mysteries, layered motives, and far-reaching consequences. Up to 16K on Pro, 48K on Ultra, and 64K on Legendary
- New model Grok 4.7 (Ultra+): complex scenes and character decisions that shape what happens next. Up to 16K on Ultra and 24K on Legendary
- Previous Luna and Grok 4.5 selections move to their replacements automatically; Grok 4.6 and the MiMo 2.5 models remain available
- Lighter page loading and smoother animated backgrounds across LoreWeaver
- Improved discoverability for public worlds and creator pages
DeepSeek V4.1 Flash & Stronger Story Memory
- New model DeepSeek V4.1 Flash (Plus+): responsive roleplay, detailed scenes, and long-running adventures. Up to 16K on Plus, 32K on Pro, 64K on Ultra, and 100K on Legendary
- Hermes 4 70B is no longer available
- Relevant lore is prioritized more consistently around the current scene
- Roleplay actions written in brackets are handled more consistently by lore recognition and Loom triggers
- Memories better preserve character identities, story dates, elapsed time, and the order of events, and keep real events apart from dreams, plans, and recollections
- Memory recall is more resilient when a service is temporarily unavailable, with fixes for relevant memories being overlooked
- Branching from an earlier message no longer carries over a summary of events beyond that point, and edits, rewinds, and regenerations are better protected against outdated summaries returning
- Fixed a sign-in issue that could stop generation, lore, Looms, and settings working after the site had been open for a while
- Temporary loading problems no longer send you away from your world or show it as missing, and you can sign in again in another tab and return to your work
- Fixed replies being cut off earlier than your selected response length
- Another reliability pass across storytelling models, the free model, and background features
Gemini 3.8 Flash, Muse Spark 1.3 & Sharper Lore Awareness
- New model Gemini 3.8 Flash (Pro+): polished prose, responsive pacing, and a steady grip on complex scenes. Up to 16K on Pro, 48K on Ultra, and 80K on Legendary
- New model Muse Spark 1.3 (Plus+): vivid character interplay and strong world-state awareness. Up to 16K on Plus, 32K on Pro, 64K on Ultra, and 100K on Legendary
- New model Hunyuan 4 Preview (Pro+): ambitious living worlds, shifting relationships, and large casts. Up to 16K on Pro, 32K on Ultra, and 48K on Legendary
- Lore cards can activate when their title or character name appears naturally in the story, even without a matching tag
- Lore recognition follows the current scene further back, so characters stay consistent as conversation moves from names to pronouns
- Player personas are preserved more reliably in context-heavy stories
- When Loom directions involve specific characters, their lore is prioritized without pulling the whole cast into the scene
- Faster recovery from temporary availability and capacity problems, including for DeepSeek V4 Pro and Hunyuan 4 Preview
- Recommended model selections refreshed around speed, reliability, and storytelling quality
- Ox Alpha has graduated into GLM-5.3 Flash, and existing selections move over automatically
August 2026
GLM-5.3 Flash, Qwen 3.8 Flash & Clearer Errors
- New model GLM-5.3 Flash (Plus+): frontier-level planning at a faster pace, with excellent continuity and precise direction
- New model Qwen 3.8 Flash (Plus+): exceptionally responsive, with crisp scene logic and coherent handling of complex plots
- Both offer up to 16K on Plus, 32K on Pro, 64K on Ultra, and 100K on Legendary
- Recovery from a wider range of temporary model-service and capacity problems, including across the DeepSeek V4 family
- Aion 3 handles slower starts more patiently and can retry without silently switching models
- Blocked scenes now show a clear explanation instead of a misleading provider-unavailable error
- Fired Loom directions are followed more consistently across models, and safeguards keep Loom instructions out of story prose
Saved Regenerations, Compact Lore & Loom Folders
- Regenerated responses are saved instead of replacing the previous version, with a navigator beneath the message to move between them
- The version you select becomes the story’s path forward, and memory, summaries, and Emergent Lore stay aligned with it
- Smoother regenerate and undo transitions, and a failed regeneration leaves your existing response in place
- Long, complex stories keep relevant lore, recent events, and continuity in view more intelligently, with stronger protection against drift
- Loom directions stay behind the scenes, with stronger safeguards against them appearing in the story
- A Compact layout for the Lore tab alongside Gallery fits far more lore on screen, and your choice is remembered
- Loom folders organize threads without changing how they activate, and survive publishing, updates, downloads, duplication, imports, and exports
- Delete all threads, with a confirmation first; removing a folder never deletes its threads
- Optional message bubbles for AI responses, with bubble colors and separate opacity for your messages and the AI’s
Faster, More Dependable Generations
- Automatic recovery when a model service is overloaded, unavailable, or slow: stalled requests move to a healthy alternative before a reply begins
- Long-running stories prepare for generation more efficiently, so responses start faster during busy periods
- The free model now has several independent recovery paths, and unlimited free-model messages are unchanged
- Nemotron 3.5 Lightning recovers more reliably when its usual service is overloaded
- Memory processing, Emergent Lore, starter actions, lore descriptions, world creation, and image prompts run through a faster, more resilient layer with a backup when needed
Referral Rewards & Backgrounds for Every World
- Referrals: every member has a personal referral page and invite link. When a friend joins and gets 10 successful story responses, you both unlock a Deep Context Day
- Rewards add +12K context for referrals one through four, +18K for the fifth, +12K for six through nine, and +24K for the tenth, then the cycle repeats; invited friends get their own +12K day
- Rewards add context beyond your plan’s limit on every model you already have, including the free model, and can be saved and activated whenever you choose
- Notifications and emails when someone joins through your invite and when a reward unlocks, with progress on your referral dashboard
- 16 curated background themes in Gradients, Auroras, and Atmospheres, including Velvet, Blood Moon, Neon Rain, Gilded Aurora, Loreweave, Starlight, and Eclipse
- Eight story fonts, including Instrument Sans, Instrument Serif, Literata, and Atkinson Hyperlegible, plus adjustable reading width and line spacing
- Custom backgrounds for any world, with automatic optimization, reading contrast, focal-point controls, and desktop and mobile previews
- Creators can include a background when publishing, and downloaded, duplicated, imported, and exported worlds keep theirs
- Follow-up on August 24: DeepSeek V4 Pro context set to 48K on Ultra and 64K on Legendary
A Smarter Model Picker & Million-Token Storytelling
- A redesigned model picker with a curated Recommended selection for each plan, covering character writing, complex plots, dark fiction, atmosphere, and speed, plus Advanced for the full catalog
- Every model has a practical description of how it performs for storytelling, and plan requirements, context, and locked options are easier to read
- Ox Alpha (Plus+): an experimental model with up to 1 million tokens of context on every paid plan while in Alpha
- LoreMap search across titles, categories, tags, and descriptions, jumping straight to the node
- Loom’s thread list now scrolls independently from the IF/THEN editor
- Fixed some users being told they had reached a daily limit while using the unlimited free model
- Context rebalanced for Gemma 4 31B, Hermes 4 70B, LongCat 2, and DeepSeek V3.2
- Follow-up on August 22: Gemma 4 31B up to 48K on Ultra and 64K on Legendary, Hermes 4 70B and LongCat 2 up to 48K on Ultra and 80K on Legendary, DeepSeek V3.2 up to 48K on Ultra and 64K on Legendary, and DeepSeek V4 Pro up to 64K on Ultra and 100K on Legendary
Nine New Models, Safer Editing & Stronger Looms
- Nine new models: GLM-5.3 (Ultra+), Qwen 3.8 27B (Pro+), Gemini 3.7 Flash (Pro+), Qwen 3.8 2.4T A95B (Ultra+), Grok 4.6 (Ultra+), a new DeepSeek V4 Pro revision (Ultra+), Seed 2.1 Turbo (Pro+), Nemotron 3.5 Lightning (Plus+), and Muse Spark 1.2 (Ultra+)
- A clear Save now button beside autosave in world settings and Looms, and failed saves keep your changes for another attempt
- More dependable saving on mobile, including after switching apps or returning to a suspended tab
- Long lore cards no longer disappear after generating and saving an AI description, and new cards use the category you are browsing
- Loom enable and disable actions now work for single threads and groups, and Looms can add facts to persistent memory or replace the remembered state
- Blocking a creator hides their worlds across the Hub, Following, search, and suggestions
- Refined Hub search, filters, content-rating controls, and mobile navigation, plus polish for creator profiles, dashboards, notifications, ratings, the image library, and story scrolling
- Upgrades and renewals recover more cleanly from past-due or incomplete payments
A New Visual Identity, Showcases & Unlimited Looms
- New models: DeepSeek V4 Flash Preview (Plus+) and Inkling Small (Pro+), each up to 100K context on Legendary
- A completely new visual identity: the landing page and public pages rebuilt around a quieter, richer, more cinematic design language
- Plans, How It Works, genre pages, sign-in, legal pages, and a refined Feature Guide brought into the same design
- Publishing rebuilt: covers, descriptions, ratings, story previews, and shared lore reviewed together before you publish
- Optional Showcases for published worlds, with their own theme, typography, introduction, quotes, images, videos, and story sections, and live previews
- Faster, more resilient memory recall, and more dependable Emergent Lore with better duplicate protection
- Loom threads are now unlimited on every plan
- Image generation can recover through another service when the first one is unavailable
- The Content Guidelines now apply to every publicly visible image: mature themes remain supported, but public imagery stays non-explicit
July 2026
Three New Models, Faster Generations & Safer LoreMaps
- New model Gemini 3.6 Flash (Ultra+): precise, polished, and good at detailed direction. Up to 16K on Ultra and 24K on Legendary
- New model Gemini 3.5 Flash Lite (Pro+): a lighter, quicker Gemini. Up to 16K on Pro, 32K on Ultra, and 48K on Legendary
- New model LongCat 2.0 (Plus+): built for long-running logic. Up to 16K on Plus, 32K on Pro, 64K on Ultra, and 100K on Legendary
- A full reliability pass on every model, with fewer dead responses and endless waits
- A subtle animated line now marks where the next passage will arrive while a response forms
- Refreshing a LoreMap safely adds missing lore cards, and maps no longer show fewer nodes than cards
- Deleting a Loom thread now removes the HUD elements it created
- Gemma 4 26B retired; Gemma 4 31B remains
Inkling, a Leaner Picker & a Dialogue Fix
- New model Inkling (Ultra+): the debut model from Thinking Machines, with custom storytelling prompts from day one. Up to 16K on Ultra and 32K on Legendary
- GLM-5, Hermes 4 405B, and Mistral Large retired
- Closing quotation marks no longer disappear when a response is trimmed mid-dialogue
Kimi K3 & Memory Fixes
- New model Kimi K3 (Ultra+): distinct character voices and vivid detail without the purple. Up to 12K on Ultra and 16K on Legendary
- Memory and summaries now follow the timeline you kept, ignoring rerolled responses and rewound scenes
- Stopped generations no longer leak into memory
Five New Models & Bug Fixes
- New model GPT-5.6 Luna (Ultra+, beta): fast, precise, and good at doing exactly what you ask
- New model Grok 4.5 (Ultra+): sharp scene logic with a strong emotional read
- New model Aion-3.0 (Ultra+): the flagship dark-fiction engine, with relentless tension and consequences that stick
- New model Aion-3.0 Mini (Pro+): the same dark instincts, tuned for speed
- New model Hunyuan 3 (Plus+): steady continuity and clean prose, up to 100K context on Legendary
- Loom dropdowns open upward when there is no room below, so lists no longer get cut off
- Generate with AI on lore cards and Emergent Lore suggestions work again
June 2026
Loom Engine 2.0
- Brand-new Loom editor rebuilt to match the story view — bottom-sheet pickers, real sliders, non-clipping dropdowns, and motion throughout on desktop and mobile
- Every condition, action, and setting now has a tooltip, and each thread shows a plain-English summary of what it does
- Adventure HUD: a live floating panel showing every stat, quest, and item the player is carrying — updates the instant a thread fires
- Meters: bind a counter to a labeled bar like Health, Sanity, Reputation, or Affection
- Quests: an objective log that ticks off as you play — each objective can be active, complete, or failed
- Inventory: items are now real, with quantity and description — plus new "has item" / "lacks item" conditions to gate what happens next
- Item, objective, and key interactions get woven into the AI's writing so the scene reacts to what actually happened
- Tappable choices: an action that drops real clickable choice buttons into the chat
- Manage panel with Flags, Items, and Quests tabs listing everything defined across your world with usage counts
- 43 one-click templates spanning fantasy, horror, sci-fi, romance, mystery, survival, combat, character chats, and director tools — now searchable by genre
- Complete Loom Engine guide rewritten top to bottom for 2.0
Every Model Rewritten, GLM-5.2 & Yearly Plans
- Ground-up overhaul of how every model writes — each one tuned to its own strengths, voice, and instincts
- Prose now follows your lead — first or second person, past or present tense, mirroring you instead of forcing a format
- New model GLM-5.2: deepest narrative reasoning in the line, available on Ultra (16K) and Legendary (24K)
- Yearly plans on every tier — commit for the year and get two months free, with effective monthly rate and savings shown on each card
- Fairer plan changes: switch mid-cycle and get properly credited for time already paid
- You always land on your plan's best available model, and the app tells you if it moves you
- Subscription engine hardened for reliable renewals, plan changes, and the new yearly option
- Fixed LoreMap node counts capping at 500
Redesigned Lore Cards & Constellations
- Redesigned lore cards rebuilt from scratch with better spacing and clearer type
- Empty states are now generated constellation art, unique to each entry and tinted to its type
- Constellation art now appears everywhere an image is missing — My Stuff, the Hub, creator profiles, published world pages, and the publish flow
- My Stuff refreshed with the new frosted-glass card texture
- Sort menu now sits above world cards, and card footers stay pinned to the bottom
- Smoother story loading with proper skeleton states; refreshed branch view and onboarding touch-ups
Stop Button & Rewind
- Stop button halts generation the moment you press it — no more waiting on a reply you already know is wrong
- Rewind to here: every AI message can roll your story back to that exact point
- DeepSeek V4 Pro context tuned to 24K (Ultra) / 48K (Legendary) to keep it fast and reliable
- Polished starter pills, corrected the tone/genre character limit (255), and fixed phantom "recently updated" worlds
The Story Experience, Rebuilt
- LoreWeaver now feels native on mobile — a glass sidebar holds navigation, New Session, tools, and your account
- A top bar that stays out of the way: just the menu, your title, and your profile
- Lore search, Memory, Generation, and Appearance now open in a drag-to-dismiss Tools sheet on mobile
- A unified glass composer: action type, model picker, actions, and text field in one frosted capsule on desktop and mobile
- Organize lore your way — drag to reorder cards (it sticks), plus a Sort control that never overwrites your layout
- Tags now show on each lore card on a single line
- Story search rebuilt from scratch as a frosted spotlight-style panel
- Font size is now a real five-step slider in Settings and the story sidebar
- Cleaner session drawer with a sleek popover menu (Edit, Branch, Export, Clear, Delete)
- Nicer Plans page with chevron navigation and your full profile up top
- Playtest copies can no longer be turned into publishable worlds
Hub Carousels, Story Export & MiniMax M3
- Real carousel arrows on every Hub scenario row on desktop, with smooth edge-aware scrolling
- Genre tag reimagined — now inline with rating and stats at the bottom of each world card
- Export chat: download any session as a clean, formatted file with every message labeled by speaker
- Loom thread filter no longer caps at the first 10 tags — every tag shows up and filters properly
- New flagship model MiniMax M3: deepest emotional range yet — Pro 32K / Ultra 48K / Legendary 64K
- More models opened to lower tiers with bigger context: MiMo-V2.5 (now Plus), MiMo-V2.5 Pro (now Pro), Qwen 3.6 Flash (now Plus)
- Retired Claude Haiku 4.5 and MiniMax M2.7
May 2026
Update Diff Viewer & Redesigned Editors
- Full diff viewer for creator updates — your version side by side with the creator's, with changed words highlighted inline
- Updates are now fully additive: lore cards and Loom threads you created yourself are never touched
- Changes grouped into new, updated, and removed cards for both lore and Loom threads
- Fixed worlds incorrectly showing "Update available" for changes that don't affect your copy
- Completely redesigned lore editor — spacious two-pane on desktop, drag-to-dismiss bottom sheet on mobile
- Completely redesigned Settings, split into categories with a sidebar, mobile tabs, and search (Cmd/Ctrl + K)
- Persona character limit raised from 3,000 to 5,000, plus live character counts across fields
Playtest Links & LoreMap Export
- Generate a playtest link from My Stuff — each captures a frozen snapshot of your world (lore, Loom threads, LoreMap, cover)
- Testers sign in, accept the invite, and get a private copy they can play and edit but not publish, export, or re-share
- Create a new link anytime to share an updated snapshot, or revoke links whenever you want
- Export your LoreMaps directly from the LoreMap toolbar
- Fixed suggested actions always pulling from the first session starter instead of the active one
Hub Update Pills & Color Wheel
- Downloaded Hub worlds now show an amber "Update available" pill when the creator pushes changes
- Pulling an update keeps your sessions, conversations, memories, custom title, cover, and audio preference — only creator content is replaced
- Hub surfaces the Discord invite more prominently
- Regular text, dialogue, and action colors now use a full color wheel with brightness and hex input, from the sidebar or Settings
Context, Hidden Lore & Loom Reads AI Replies
- DeepSeek V4 Pro context up to 32K (Ultra) / 64K (Legendary)
- Hidden lore cards in the Context Inspector render as locked tiles instead of opening the edit drawer; inline tooltips respect the hidden flag
- Editing one world no longer re-stamps every world's updated date — "Last played" sort stays honest
- Loom threads now fire on keywords in the AI's last message, scanned exactly once per turn
Qwen3.7 Max & Memory Fixes
- Added Qwen3.7 Max to the model roster
- Session Recap is now a tight "story so far" paragraph instead of a briefing-doc bullet dump
- Memories record what literally happened — no inferring, upgrading, or parroting your world bible
- Your live conversation always wins; recap and memory bank are reference only
- Recap stays out of early sessions until there is actually story to recap
- Duplicate-memory prevention, more reliable auto-summarization timing, and tighter compression
Memory System Rebuild
- The entire memory system rebuilt from the ground up
- Auto summaries are dramatically smarter and pull more insight before generating
- Memories understand your scenes, so character continuity, plot threads, and world details carry forward cleanly
- Long sessions in particular should feel night and day
- Smart image generation for lore cards is way more reliable
Grok 4.1 Fast Removed & Smarter Plumbing
- Grok 4.1 Fast removed (xAI pulled it) — anyone who had it selected was auto-routed to DeepSeek V4 Flash
- AI world generation and image prompt generation upgraded to smarter models for sharper, more consistent results
April 2026
Qwen 3.6 Flash, Grok 4.3 & Owl Alpha
- Qwen 3.6 Flash: Pro 32K / Ultra 64K / Legendary 100K (replaces Qwen 3.5 Flash)
- Grok 4.3: Ultra 16K / Legendary 24K (replaces Grok 4.2)
- Limited-time test run of Owl Alpha — a stealth model with a 1M context window, free for Plus and up
- Removed MiMo V2 Omni and MiMo V2 Pro
- Fixed internal IDs showing instead of lore card titles, cleaned up incorrect icons, and fixed the mobile search page
DeepSeek V4 Flash & Pro
- DeepSeek V4 Flash: fast long-context DeepSeek — Plus 16K / Pro 32K / Ultra 64K / Legendary 100K
- DeepSeek V4 Pro: premium reasoning and prose — Ultra 12K / Legendary 16K
- Removed Elephant Alpha and Euryale 70B
MiMo-V2.5 & MiMo-V2.5 Pro
- MiMo-V2.5 Pro: architect of intricate plots — Ultra / Legendary 16K
- MiMo-V2.5: tight plot logic and clean exposition — Pro 16K / Ultra / Legendary 32K
- Older MiMo 2 variants scheduled for removal
Kimi K2.6 & Claude Haiku 4.5
- Kimi K2.5 replaced with Kimi K2.6
- Claude Opus 4.7 removed (test run complete)
- Limited-time test run of Claude Haiku 4.5 — Anthropic's fast, efficient small model
Claude Opus 4.7 (Test Run)
- Limited-window test run of Claude Opus 4.7 — Ultra 8K / Legendary 12K
LoreMaps 2.0
- LoreMaps rebuilt from the studs and now free for every user on every tier and every world
- A completely new look: smooth curved connections, color-coded tagged relationships, softer shadows, and smoother motion
- Drag anything, anywhere — nodes are fully draggable and your layouts save automatically
- Your relationships now shape the story: tagged connections (Family, Rivals, Sworn Enemies) can pull the right lore into context automatically (opt-in per world)
- Publish your LoreMap to the Hub along with your world, context setting included
- Finally works on mobile — pinch to zoom, drag, and slide-up panels
- Dramatically smoother panning, zooming, and dragging even on huge maps
- All existing maps preserved exactly as you left them
- Expanded markdown support: headings (#, ##, ###) now render in stories
- Fixed users being briefly switched to Free tier near subscription renewal
Free Tier Boost & Dynamic Response
- Free tier premium messages increased from 5 to 25 per day
- Dynamic response length option — the AI adjusts how much it writes based on what the scene calls for
- Multiple session starts per world — create different starting scenarios without separate worlds
New Model: Elephant Alpha
- Stealth beta model Elephant Alpha: efficiency-focused 100B with strong reasoning and minimal token waste
- Context: Plus 64K / Pro 128K / Ultra / Legendary 262K
Qwen 3.6+ Replaces Qwen 3.5+
- Qwen 3.5+ replaced with Qwen 3.6+ — significant quality upgrade with new tuned sampling, now from Pro and up (16K/32K/48K)
- MiniMax M2.7 context increased to 48K for Legendary users
New Model: GLM-5.1
- GLM-5.1 available from Ultra and up at 16K context
- GLM-5 context increased to 24K for Legendary users
- Removed Mercury 2 from the model lineup
New Models: Google Gemma 4
- Google Gemma 4 26B and Gemma 4 31B available from Plus and up, scaling per tier (16K/32K/64K/80K)
- Fixed "See All" on Hub categories not applying the correct filter
- Fixed Hub, created, and download filters on My Stuff not persisting
Bug Fixes & Grok 4.20 Beta
- Grok 4.20 Beta working again — Legendary users now get 12K context (up from 8K)
- Prism Ultra now supports up to 80K context for Legendary users
- You can now delete memories from the Settings tab
- Settings tab UI updated to better use space on larger screens; Plans page makes Legendary easier to find
- Fixed soundtrack player overlap issues and hidden lore card details being viewable through the LoreMap
March 2026
Block Creators & Threaded Comments
- Block creators to hide their worlds from the Hub and search, managed from your creator page
- Threaded comments — reply to comments on published worlds with full nested replies
- Redesigned published world edit page with editable lore card tags
- Showcase is now optional and disabled by default when publishing
- New showcase theme: Void
- Fixed hidden lore cards being visible in the sidebar and cut-off Loom sub-condition dropdowns
Session Notes
- The Character tab is now "Author" and includes a free-form Session Notes field, weighted strongly in context
- New Loom thread THEN type: Inject into Session Notes based on story triggers
- Fixed italic/bold inside dialogue breaking dialogue color formatting and bold always rendering white
- Fixed being unable to add more than 2 images at a time to a world's showcase
LoreMaps, Publishing & Sidebar
- Fixed jarring zoom when creating or deleting LoreMap edges; added a "Delete All" button and a panel to view all nodes and relationships
- Downloaded worlds can no longer be cloned or exported, protecting creators
- You can now hide premise and tone/genre from users who download your world
- New My Stuff toggle between created and downloaded worlds
- Sidebar no longer shrinks your story area unless it runs out of space; edit lore, memories, and summary directly from it
- Overhauled world duplication — faster and correctly copies all Loom threads
Desktop Story Sidebar
- Desktop sidebar in the Story tab — view lore cards and memories and adjust settings without leaving the story
- On desktop, story text takes up more space rather than being restricted to one column
- Rewrote all model descriptions in Settings to better explain storytelling performance
- Hub search now respects your content rating preferences
- Fixed markdown code block formatting to wrap properly
Guidelines & Context Budgeting
- Revamped community guidelines with a clear breakdown of what is and isn't allowed and how to rate your world
- Updated the feature guide with new publishing info
- Optimized context budgeting for lore-heavy worlds so significantly more lore cards fit
Color & Font Customization
- New color and font options in Settings: story font, regular/dialogue/italic action colors, user message bubble color, and bold dialogue toggle
- Showcase media limit increased from 6 to 24
- Added a confirmation modal when branching worlds; fixed mobile quick actions going off screen
Moderation & Content Ratings
- Content moderation framework — worlds with illegal content are blocked from publishing
- Age verification prompt for all users (18+)
- Content rating filter on the Hub: Everyone, Teen, or Mature (defaults to Teen, Mature is opt-in)
- Lore card, premise, and style rules context budgeting handles overflow more gracefully
- Major performance improvement downloading worlds with lots of lore card images; tuned POV handling
Markdown & MiMo-V2 Models
- Story text now renders underline, inline code, and code blocks
- Image Style Switcher: when See is selected, the model switcher becomes a realistic/anime style toggle
- MiMo-V2 Omni: omni-modal agent model from Xiaomi — Pro 16K / Ultra+ 32K
- MiMo-V2 Pro: 1T MoE frontier model (formerly Hunter Alpha) — Ultra 16K / Legendary 32K
- Retired Hunter Alpha and Healer Alpha (now official MiMo-V2 models)
- Max persona/player context increased to 3,000 tokens
Creator Announcement Board & Publishing Overhaul
- Animated Creator Announcement Board on your Creator Profile
- Revamped publishing flow — configure lore cards, style rules, images, and descriptions from one streamlined page
- Mark specific lore cards and style rules as hidden from downloaders
- Copy the text of any user or AI message with one click; view your full follower list
- New models: Mistral Small 4 (Plus+, 16K/32K/64K/100K) and MiniMax M2.7 (Pro+, 16K/32K)
- Better TTS emphasis handling, fixed lore card context budgeting, any-aspect-ratio showcase images
- Downloaded worlds can no longer be re-published by someone who isn't the original creator
Last Played Sort & Formatting
- New "Last Played" sort in My Stuff by most recent session activity
- Style Rules character limit increased from 8,000 to 12,000
- Bold and italic formatting now renders correctly inside quoted dialogue
Model & Performance
- Hunter Alpha and Healer Alpha no longer count toward daily premium message limits on any tier
- Backend performance improvements for long stories
- Updated content filter for published worlds — relaxed overly strict triggers, better coverage for harmful content
Persona Images & Gallery
- Personas now support images — upload with a crop tool or generate with AI from the Personas tab
- Persona image displayed in the Player Character tab of your session
- New Gallery section in Settings showing all AI-generated images from your session in one place
Free Model Select & Template Variables
- Free users can explicitly select the free model to bypass premium message limits entirely
- {{user}} and {{char}} template variables supported across character and regular worlds
- Bold text formatting now supported in story output alongside italics
- Improved markdown parsing for inline formatting with better nesting support
Free Tier Model Upgrade
- Default free model upgraded from Mistral Nemo to Hermes 4 14B — smarter, more creative, better at following style rules
- Premium model daily budgets now visible for Ultra users (750 messages/day across premium models)
Creator Profiles, Showcase Builder & FictionLab Import
- New Profile Studio — upload and crop a cinematic hero banner, link socials, add a tip link, and pin your favorite world
- Showcase Builder turns your world page into a premium interactive landing experience
- 7 custom themes and 5 typography styles, Story Sections, Media Blocks (images and YouTube embeds), and a real-time Live Preview
- FictionLab story import support (best-effort) — pulls in story JSON, chat history, player character, and more
- World Description and Preview Text are now optional; sleeker dropdown menus and polished mobile layouts
Hunter Alpha & Healer Alpha
- Two new beta models with expanded context for a limited time
- Hunter Alpha: Plus 32K / Pro 128K / Ultra 1M
- Healer Alpha: Plus 32K / Pro 64K / Ultra 262K
- Text fields across world settings, lore, and summaries now auto-resize as you type
- Increased limits: World Premise 12,000, Style Rules 8,000, Lore Descriptions 20,000
Aurora Theme & Memory Ordering
- New theme: Aurora — brings the dynamic Hub/character-call background into your Story tab
- Memories now injected in chronological order instead of by relevance score for better timeline understanding
- Action counts now displayed for each memory in the Context Inspector
My Stuff Overhaul
- Search bar in My Stuff to quickly find any world or published story by name
- Sort controls (recently updated, newest, oldest, alphabetical) saved between sessions
- View mode toggle between grid and list view, also saved
- Your last active My Stuff tab is now remembered; toast notifications redesigned for mobile
Message Budgets & Notification Polish
- Premium model daily budgets now visible in the UI (Pro gets 350 messages/day across premium models)
- Notification dropdown redesigned with cleaner layout and improved readability
- GLM-4.7 Flash now correctly serves 64K context for Ultra users
- Mobile navbar in the Hub cleaned up; various mobile UI polish and clipping fixes
AI Dungeon & Novel AI Import + Theming
- Full AI Dungeon import — maps story cards to lore cards, adapts memories, brings over summary and cover image
- Novel AI import support (best-effort) bringing over all your story data
- AI Dungeon and Novel AI story cards and lorebooks can be imported into the Lore tab
- Added themes and font size options in Settings, reflected directly in the Story tab
- Fixed UI wrapping on smaller screens and a rare plan upgrade block for first-publish PLUS rewards
Story Search
- Story Search — a fast way to search through your active session and instantly jump to the exact moment
- Makes revisiting past lines, scenes, and details easy without scrolling through long conversations
New Model: Mercury 2
- Mercury 2: a premium model so fast it feels basically instant, with strong reasoning and story quality
- Available from Plus with 16K/32K/64K context for Plus/Pro/Ultra
- Refreshed model and Prism descriptions across the picker
New Logo & Lore Token Counts
- Updated LoreWeaver AI logo across the platform
- Lore cards now show a token count with color-coded indicators (yellow getting long, red trim it down)
- Editing a session now pre-selects the name so you can just start typing; mobile Hub UX improvements
Entity-Aware Memory & Mistral Large 3
- New model: Mistral Large 3 for Pro and Ultra (16K / 32K)
- Memories are now entity-aware — the system understands who, what, and where each memory is about
- Memory retrieval significantly smarter; even short actions pull in relevant scene context
- Memories categorized and ranked by narrative importance; compression is semantically grouped
- Session summaries use a structured format (characters, plot, world state, open threads)
- Kimi K2.5 moved to Ultra-only; reworked model behavior tuning and improved system prompts
Bug Fixes
- Fixed Loom threads controlling lore cards not being preserved when uploading to the Hub
- Updated MiniMax context configurations to reduce token bloat in longer stories
- Fixed world cover images not downloading along with some Hub worlds
Performance Improvements
- Major infrastructure changes — image loading, page loads, and world browsing all feel noticeably snappier
- Fixed published world lore card titles deselecting while typing
New Model: Kimi K2.5
- Kimi K2.5 from Moonshot AI — frontier-level multimodal reasoning and writing, Pro 16K / Ultra 32K
- Aion 2.0 context rebalanced to 16K (Pro) / 32K (Ultra)
- Your input is now preserved on model errors so you can retry without retyping
- Return key on mobile now inserts a newline instead of sending
Loom Engine: Full Edit Access on Published Worlds
- Edit every piece of your Loom Engine directly from your published world page
- Flag Management Hub — create, edit, delete, and view all flags in one place with thread references
- Flag interpolation in thread fields via a dropdown picker, resolving at runtime
- Direct Output Mode — threads bypass AI generation and inject your exact text into the narrative
- Loom Engine Debug panel in the Story tab showing live flag values, active threads, and last actions
- Loom Engine and Lore Card import/export as JSON
February 2026
The Creator's Update
- Full real-time notification system with a bell icon for downloads, comments, replies, likes, and follows
- Milestone achievements across views, downloads, likes, comments, and followers, plus notification preferences
- Creator Dashboard: private Overview, Worlds Management, 30-day Analytics charts, Announcements, and Settings
- Announcements broadcast system to send direct updates to everyone who downloaded a specific world
- Redesigned public creator profiles with banner images, social links, pinned world, and rich 16:9 world cards
- Follower counts and one-click Follow system
- Hub updates: Featured World section, Recently Updated carousel, rich clickable author chips, and a Share button
Loom Engine Evolution & Next-Gen Image Models
- Dice Rolls — add elements of chance with automated success/failure checks behind the scenes
- Dynamic Persona Updates — Loom actions can permanently modify a character's core persona mid-session
- Launched the official interactive Loom Engine Guide covering threads, conditions, actions, and world state
- Brand-new image generation models across the platform with a massive quality jump (the new anime model especially)
- Rebuilt Image Studio with a cleaner interface and refined style controls
- Improved world and character creation flows, especially for character creation
LoreWeaver Prism
- LoreWeaver Prism — our first proprietary AI engine that blends the best of multiple models per scene
- LoreWeaver Prism for Pro (32K) and LoreWeaver Prism Ultra for Ultra (64K) — look for the ◆ in the model picker
- Improved load performance on the Community Hub page
Grok 4.1 Fast & Smarter Backend
- Grok 4.1 Fast added for Plus and above — Plus 16K / Pro 32K / Ultra 64K
- Upgraded the internal AI powering background tasks — memories, summaries, emergent lore, and AI lore cards are sharper
New Models: Hermes 4, GLM-4.7 Flash & More
- Hermes 4 405B for Ultra (16K) — the largest open-weight roleplay model offered
- Hermes 4 70B for Pro and above (Pro 32K / Ultra 64K)
- GLM-4.7 Flash for Pro and above (Pro 32K / Ultra 64K)
- Qwen 3.5 Plus Flash (Plus 16K / Pro 32K / Ultra 64K)
- Plus subscribers now get 150 premium messages/day (up from 50)
- Qwen 3.5 Plus upgraded to the full 397B parameter model; tier-based context for Euryale 70B and GLM-4.7
- Aion-2.0 moved to Pro minimum; all model system prompts rewritten
- Plans page model/context matrix, effective context shown per model, and tier-grouped model pickers
Memory System Overhaul
- Completely rebuilt how memories are created, summarized, and recalled
- Memory generation produces significantly richer, more detailed story summaries
- Recall ranks memories by relevance to what's happening right now, not just recency
- Publish World flow updated with a cleaner dropdown and a dedicated toggle for including characters
- Main story area rendering optimized — generations feel noticeably snappier in longer sessions
Ultra Tier & Plan Restructure
- New Ultra tier: exclusive GLM-5 and Qwen3.5 Plus, unlimited memories, 64K context, priority image gen
- Ultra includes 400K TTS characters and 360 call minutes/month plus a premium gradient badge
- Restructured tiers: Free (Mistral Nemo + Trinity), Plus, Pro and Ultra
- Pro context moves from 64K to 32K and memories cap at 200; GLM-5 and Qwen3.5 Plus become Ultra-exclusive
- New model Aion-2.0: built on DeepSeek V3.2, tuned for immersive roleplay with darker themes (Plus and above)
The Loom Engine
- Loom Engine: visual scripting system for worlds and characters — define IF/THEN threads that evaluate silently every session
- 40+ condition types: message content, player input mode, action count, randomness, world state flags, lore card status, cross-thread relationships
- Composable logic with AND, OR, N-of, and nested logic groups
- Actions: narrative directives, NPC dialogue, scene transitions, lore card activation/deactivation, pin cards, set flags, increment counters, enable/disable threads, toasts, achievements, image generation queuing
- Persistent world state: key-value flag store per session for tracking game state across threads
- Full-page Loom Editor with thread list sidebar, drag-to-reorder, tag filtering, search, and fire count badges
- Live World State Inspector showing current session flags in real-time
- Publish visibility: Exposed (downloaders can see threads) or Hidden (threads run silently)
- Context Inspector now shows Loom as its own category with directive token count and fired thread names
Quality & Intelligence Improvements
- Smoother message rendering — fixed flash when streaming messages finish and save
- Characters tuned to use tone and word choice over physical action annotations for more natural writing
- Image generation triggers significantly expanded — picks up character arrivals, outfit details, physical scenes, and dramatic moments
- Emergent Lore detection improved — no longer creates duplicate cards for your own character
Soundtracks & Navigation Polish
- Soundtrack playback stability and audio quality improvements
- Navigation overhaul — profile menu now gives quick access to all pages in one place
Story Branching
- Story Branching: fork your story into a new timeline at any message — branches inherit memories, summaries, and world lore
- Redesigned Session Drawer with git-style tree rendering showing where each branch split
- Color-code sessions to keep timelines organized
- Branch Tree View: full interactive node graph of your entire story multiverse
- Branching available for all users: Free, Plus, and Pro
- Major bugfixes and fine-tuning for Character Calling
- Upgraded TTS to higher fidelity model — noticeably cleaner voices
- Hub redesign with cleaner layout and better discovery
Voice & Character Calling
- Full Text-to-Speech system powered by the #1 ranked TTS model — hover any AI message to hear it read aloud
- Character Calling: open a real-time voice conversation with your Character — talk, they talk back
- Voice library with handcrafted presets from commanding villain to whispery ASMR narrator
- Voice cloning from a 10-15 second recording (Pro)
- Monthly voice limits: Free 10K chars/10min calls, Plus 50K/60min, Pro 200K/180min + cloning
- Full published world editing — update cover image, lore entries, and more from the hub page
- AI cover art generation in styles: realistic, anime, fantasy, cyberpunk, gothic, watercolor
- Version history and changelog entries for published worlds
Auto Scene Generation, Everywhere
- Auto scene generation now works in all Worlds, not just Character Worlds
- AI reads your story and generates images when the moment calls for it — no mode switching
- Import button on My Stuff page moved to a more intuitive spot
- General stability and security improvements
- Fixed issues with some models not respecting generation settings
Auto Scene Generation & Art Styles
- Auto Scene Generation (Pro, Character Worlds): AI watches for visually significant moments and generates images automatically
- Art Style Toggle (Pro): switch between Realistic and Anime art styles in Settings
- Bug fixes for image generation backend — smoother and more consistent
New Models: GLM-5 & Aurora Alpha
- Rebuilt model infrastructure for faster model onboarding
- GLM-5: available for Plus and Pro with 16K context
- Aurora Alpha: brand new beta model, 128K context for Pro, 16K for Plus
SEE Engine & Image Chaining
- Partnered with VoyceMe SEE Engine for a massive leap in image generation quality
- Revamped See Mode (Pro): type a description or let it visualize the current scene
- Image Chaining (Pro): every generated image builds on the last — your character stays consistent across scenes
- Personas (All Users): create and manage player character profiles from My Stuff, set a default that carries into every world
- Feature Guide: full walkthrough of everything LoreWeaver AI can do, available from the Hub
Performance & Readability
- Updated Emergent Lore to fix inefficiencies and improve performance
- Fixed delayed AI message generations caused by external API issue
- Minor visual update to dialogue for better differentiation from other text
Published World Editing & Lore Images
- Edit your published world directly from the Published World page
- Edit session starter from the Settings page
- Fixed uploaded lore card images not saving
- Fixed inability to remove a lore card image without replacing it
- Cover images now update correctly after uploading
LoreMap Edge Customization
- Interact with and customize connections between lore cards — relationship tags and deletion
- New connection UI for creating and managing relationships directly from LoreMaps
- Richer interaction states: hover, select, and connect with improved visual clarity
Quality & Sustainability
- Emergent Lore now available to Plus and Pro users only for sustainability
- Updated internal system prompts for improved response quality and coherence
- Updated model context lengths in Settings UI to reflect real limits
LLaMA Euryale & Model Updates
- New model: LLaMA 3.3 Euryale 70B — storytelling-focused with stronger dialogue and scene flow
- MiniMax now runs with 32K context window for more stable responses
- Fixed DeepSeek provider configuration eliminating slow responses
- Image generation styles refined for more accurate results
- Fixed dialogue closing quotation marks being cut off
LoreMaps Polish & Free Session Limits Removed
- Updated LoreMaps configuration for more connections and better character-related links
- Quick info modal — click a node to view key lore card details without leaving the map
- Session limits fully removed for Free users — infinite adventures with no session caps
- Daily limits remain on pro model usage (Free 5 / Plus 50 messages), then free model kicks in
LoreMaps
- LoreMaps: node-graph visualization of your world — see all lore cards and relationships at a glance
- Built for large, interconnected worlds with dense lore
- Available to Pro and Plus users
- Under-the-hood performance and UX refinements for large worlds
Soundtrack Overhaul
- Completely overhauled soundtrack backend for smoother, more cohesive playback
- Music now fits story mood and tone far more accurately
- Significantly reduced random disconnects and interruptions
- New music player with current mood/genre display, play/pause, and stop controls
- Fixed context limit issue affecting Qwen3 235B and GLM 4.7
New Models: Qwen3 & GLM 4.7
- Added Qwen3 235B AI model
- Added GLM 4.7 AI model
- Continued improvements to soundtrack infrastructure
Soundtracks & Redesigned Search
- Dynamic story soundtracks that evolve in real-time with your story's tone and pacing (Pro, Early Alpha)
- Completely reworked search experience for faster and more precise discovery
- Overall UX refinements to support larger worlds and faster discovery
Stability & AI Lore Tags
- AI-generated lore card descriptions now also include auto-generated tags from story context
- Increased text input area when editing messages
- Improved Settings page autosave reliability
- General performance improvements across animations and image-heavy views
- Fixed duplicate AI responses caused by messages sending twice
- Resolved UI glitches in modals, parallax sections, and images page
AI Lore Descriptions & Dark Fantasy Style
- AI-generated lore card descriptions — AI pulls context from your story to generate rich descriptions
- New Dark Fantasy image generation style for moodier, high-contrast visuals
- Switched to a new image generation model with better prompt accuracy
- Lore generation now better understands narrative tone and world context
January 2026
Context Accuracy & Character UI
- Updated Context Inspector and backend for accurate token counts sent to API and displayed
- Character Creator and Settings now use proper character-specific wording
- Fixed session starter being cleared when clearing session messages and memories
Settings Overhaul
- Total overhaul of the Settings page for better UX and accessibility
- Backend context management now handles player character details better
- Added toggle in Settings to disable autoscroll
- Fixed Emergent Lore toggle not visually updating
- Fixed edge cases causing settings to save infinitely
Mobile Performance & Trinity Large
- Context Inspector UI cleaned up to prevent text wrapping on longer stories
- Added total story length token count to Context Inspector
- Major mobile performance improvements for smoother scrolling
- Added Trinity Large model with 128K max context (experimental)
- Hermes max context lowered to 16K (actively searching for cheaper providers)
- Fixed story summary token length to prevent truncations
Character Creation Improvements
- Added Create Character option for faster setup
- Updated Character Creation flow for speed
- Image generation now pulls from character descriptions for more accurate visuals
- Fine-tuned character creation model for better quality responses
Character Defaults & Templates
- Character worlds now default to MiniMax M2-Her as intended
- Added Character template option to the world creation flow
- AI world generation now supports Character creation
- Strengthened validation on world publishing
Characters & MiniMax M2-Her
- Characters: a new world type built for one-on-one roleplay conversations with full lore, memory, and all features
- MiniMax M2-Her: a roleplay-first model available to all tiers, default for Characters
- Free and Plus users get daily premium messages with MiniMax before switching to DeepSeek
- Model switcher now available to all tiers (3 models to choose from)
- Sliders now accept direct text input for precise control
- Asterisks now render as italics to support roleplay formatting
- Fixed action type resetting when switching tabs
- Fixed player input scroll behavior
Publishing Validation
- Added validation checks when publishing worlds to ensure quality standards before going live
Quality of Life & Image Improvements
- Settings now auto-save — no more wondering if changes stuck
- Story tab layout refresh with model selector repositioned for more story space
- Context Inspector auto-scrolls to show the latest
- Image generation tuned up with noticeably better visual output
- Fixed 404 on Guidelines page
Emergent Lore & Community Features
- Quickplay — jump into any world instantly
- Creator profiles — see all worlds from a creator in one place
- Emergent Lore: every 12 actions, a standalone AI scans your story for new characters, places, and items
- Emergent Lore generates lore cards automatically with confidence scores — you approve or deny
- Your world builds itself while you play
- Bug fixes and UI polish across the board
Launch
- Memory System: keyword extraction pulls relevant memories from your entire session, not just recent turns
- Memories scored by relevance, not just recency — NPCs remember you hundreds of turns later
- Smart Summaries: rolling compression preserves critical plot points, character states, and unresolved threads
- Context Assembly Overhaul: dynamic budget allocation between lore, memory, summary, and recent messages
- Adapts based on your model's context window with relevance-scored lore injection
- Production infrastructure, proper error handling, mobile support
