We worked hard on improving the experience on slow models
World authors will also notice much better debugging
Misc:
Introduced a few facilities for solo characters (worlds where the objective is a simple setting and a single NPC):
Solo Characters and Customized characters are still not the focus of Imagina. They are not chatbots either, they are still world with trackers, etc.
Author pages are here! They are accessible via a stable URL that you can share.
Authors can write a bio, set an avatar and upload a banner. Worlds can be showcased.
Other users can follow the authors. For now, this doesn't send any notification, but we'll work on that soon!
We will keep improving them in the next weeks.
We cleaned up the image generation:
We're very excited about this as Anima can be prompted in natural language AND booru tags. This makes image generation in journeys much better and less random. We're also hosting this ourselves, meaning that no cloud provider knows about your generations. In exchange this means that this can sometimes be down.
{{user}} macro now works as an alias for <<player_name>>, and in many more fields (NPCs, trackers, endings, intro, and more), so character cards brought over from other apps keep addressing the player by name.Introducing a "new" model: Mantle Lumen!
You can now pick between 3 fonts: Standard, Serif, Monospaced.
Added a sidebar on the journey play area, to easily switch between journeys without going back home.
Turn edition! You can now edit the reply the AI wrote. Only the last turn is editable, rollback or fork if you want to edit a previous one.
We added back Xiaomi MiMo Pro as we managed to get it to work. Our default model has been Gemini 3.1 Flash since the beginning, but unfortunately Google made 3.5 too expensive for the quality. To prepare for 3.1's removal, we did a couple of things:
Unfortunately the side effect is that games are much slower to play using DS4. But we believe that the very small cost more than makes up for it.
We hope that it's the last time we have to do that. Our goal is still to have a great experience using community finetunes that will not go away.
We now have a Victory/Defeat system! Add your conditions and the game will resolve itself once met. Players can decide to continue the game. There is also an "ending" concept where you can define multiple Good/Bad endings.
A new "Hybrid" mode is now the default: Enter sends on desktop, but not on mobile (where accidental taps are common). You can switch to "always Enter sends" or "always Shift+Enter sends" in Settings.
You can now tap multiple suggestions to combine them, then tap the last one again to send. No need to manually hit Send after building up your action.
The home layout has been tweaked:
Hope you like it!
You can now copy worlds for those who enabled this option! This allows you to edit worlds and add your own twist on them. Lineage is recorded so we will be able to make a better UX in the future, showing where a world comes from and credit the original author(s).
You can now browse all AI-generated images from your journey in one place. Look for the photo icon in the top bar (or "Image gallery" in the menu on mobile).
As Delve Lumen's low cost is now permanent, is has now been promoted as a recommended model and is eligible for free turns. Delve Spark is now hidden: games using it will continue to work, but it didn't make sense to keep a weaker model when the better one is quite cheap.
You can now manually ask for an image generation using the "..." button on AI turns.
Couple of improvements:
Added "author notes" in some places. That's for you, world builders!
Added global trackers, so that you don't have to duplicate them between all of your NPCs
Moved stuff around and improved mobile support again
Added a story audit feature where an AI gives you input about how your well story uses the engine and incoherences. Very early beta, free to use!
Bug fixes for world generation with Claude. For the time being, Gemini is the default world generator.
A few new authoring tools for Loom worlds:
As the credit pricing model seems to turn off a lot of people, we added a button to show interest in a subscription model.
The world editor's UI has been rethought on mobiles:
Desktop users got a couple of benefits from this too.
But the most important information is that world edition is now possible on phones, even small ones!
Worlds can now define known locations! When location tracking is enabled, the engine will read pre-made locations from the world definition and will be able to dynamically create or update locations for better story consistency.
Location tracking is enabled by default on new worlds.
Gemini 3.5 Flash is available for texting. While it's much faster than Gemini 3 Flash, it's 2.5x more expensive in our testing. We're evaluating which model should become the default, as we believe that Google will soon stop supporting the current one.
Every published world now has a short share code (e.g. K3M-JP2) for easier sharing.
While Imagina's Spark engine has served us well during this beta period, it had a couple of limitations due to its architecture.
Loom is a rewrite that turns Imagina into an agentic story teller, where the AI can decide to loop and do work before writing the narration. This work includes consulting the world definition, past turns, check on past and present NPCs, etc.
This unlocks new possibilities for Imagina:
We initially thought that this would be too costly to run, but we managed to optimize it enough to make it worth it.
More model choice is what we believe is the best outcome. The new engine now supports models like Xiaomi's, DeepSeek and in the future community fine-tunes of smaller models. For you, this means that you will have more choice in what AI models are available: broader range of tone/behaviour and cost ranges!
This brings us even closer to our ultimate goal of breaking free of the big cloud providers and hosting our own models. No more surprise deprecation, no more tone change or storyteller you like going away.
While we will continue to support Spark for a while, after the initial Loom test ends all worlds will use it by default. This is the version we will iterate on from now on. To celebrate, we bumped the number of daily free turns from 2 to 3, passing the cost savings to you!
Hope you enjoy it as much as we do!
Each player action bubble now has its own ⋮ menu with Edit/Replay this turn. Choose it on any past turn and the story rewinds to right before that action: subsequent turns are deleted, and the action and steering fields are repopulated with what you wrote, ready for you to tweak the wording (or change your mind entirely) and resubmit. A confirmation dialog explains what will happen before anything is deleted.
The existing rollback option on AI narrations was renamed to Resume story from after this point so it reads more like what it actually does.
Reworked how prompts are built for the SDXL anime image models (Nova Anime XL, WAI Illustrious SDXL, CyberRealistic XL). Turn images, character portraits, world title cards, and editor backgrounds now run through a shared post-processor that injects a polished quality stack and a long anti-failure negative bundle on every generation. The most visible improvement: skin no longer drifts into the milky wax / melted plastic look that occasionally appeared on close-up shots. Other recurring artifacts (rainbow hair, stray extra figures, deformed hands, watermark text) are also suppressed automatically.
When a moment intentionally calls for wet skin (rain, hot spring, a sweaty intimate scene) the anti-oil bundle steps aside so the image actually looks wet. Mirror shots now get the right tag combination so the character interacts with the mirror instead of treating it as background decor. Group shots with three or more named characters fall back to a tight close-up of the most iconic subject or a peopleless scene shot of the location, since SDXL cannot reliably render that many bodies in one frame and was producing fused-limb images in those cases.
Shared world links (/world/<id>) now arrive with the world preview, name, description, tags, content warning, and portrait already in the page rather than waiting for the app to fetch them after load. Search engines and link previews see the actual world copy instead of an empty shell, and the page feels snappier on a cold open.
Every account now gets a small pool of free turns each day, usable on most models (the premium Astral Lumen tier still costs credits). Free turns reset at midnight UTC and are consumed automatically before your credit balance whenever you play an eligible model. The credits popover in the play header now shows "Free turns remaining" alongside your balance, and entries in your turn history are marked when they were covered by a free turn.
If you're on a model that doesn't accept free turns and run out of credits, sending an action will offer to switch to a free-turn eligible model instead of failing.
Every journey character now remembers each portrait you've generated. Open the character menu and click Portrait history to scroll through every portrait, with the turn it was made on and when, and tap any image to view it full screen. The one currently shown in the character sheet is marked Current.
The character sheet and fullscreen portrait viewer now show which turn your portrait was generated on, so it's easier to tell when an older image no longer matches the story.
Rolling back a journey now also rolls back any portraits generated on the discarded turns. If you had the right portrait before a rollback, you'll need to generate a new one, or the previous journey-wide portrait (if any) resurfaces.
Every account now gets one free world generation. Head to the world editor, describe a concept, and your first generation runs without costing any credits. A note on the generation panel will remind you while the freebie is still available, then disappear after you've used it.
When AI-generated turn images are enabled on your Profile, open the character menu in the play header (your character name) and use Generate portrait to create a portrait from your appearance. Credits apply like other image generation.
From Admin → Game debug, Open in Player now loads that journey correctly. The play UI is read-only for another user’s game (no actions, branching, rollback, model change, or credit shortcuts). Debug and export still work.
In Story Beats, condition rows now have move up / move down controls so you can change evaluation order (conditions still combine with AND semantics).
Worlds can now drive pacing with story beats: authored cues like "the storm finally breaks", "the rival makes her move", or "someone notices the player's pendant". The engine evaluates beats every turn — some fire on their own when their conditions match (turn count, tracker values, after another beat, a chance roll), others stay armed and let the AI decide when the moment lands. Fired beats can drop a one-shot nudge into the current turn or quietly add lasting context the AI carries forward.
You don't have to do anything — beats just show up in the story when their world uses them.
World authors will find a new Story Beats section in the world editor where the old Clock / Time lived. The clock system has been retired; existing worlds keep loading, and you can rebuild any timing rules as beats with the new condition kinds.
Worlds can now illustrate their biggest story moments: first meetings, dramatic reveals, scene changes. The AI picks when. Character-focused shots float beside the narrative, wider scenes sit above it. Each image costs about 7 credits on top of the turn.
You stay in control. Turn images off for every journey on your Profile, or disable them just for the journey you're in from in-game Settings.
World authors can opt in from the world editor under Advanced Features → Turn Images, add optional guidance for the AI, and tune how often images appear.
Full details in the help article.
/ in the action input to see available commands. /retry re-generates the last turn, /undo rolls it back, /steer <text> pre-fills the steering box, /1 through /4 quickly pick a suggested action, and /debug toggles the debug panel./world/:id) you can copy straight from the address bar. Social previews (Discord, Twitter, etc.) show the world's portrait, name, and description.