New Text Model
- Gemini 4 Argon has been added. 50% off during Google's promotional period.
- Mimo v2.6 Pro runs 9x faster. Same pricing as before.
What’s New
Mana Updates
New features, improvements and fixes shipped to Mana, newest first.
Latest update
New feature: Animate
Animate an in-game image into a 5-10 seconds clip fitting your story context.
The animate feature works both for SFW and NSFW images.
Animate is available to only paid users. Video generation takes 0:30-10 minuets depending on durations and provisioned GPU type. Animate runs in the background - allowing you to continue playing as you wait.
New Image Model: Mana
Meet Mana, our new image model for creating characters and custom in-game images.
Creating character: Mana understands natural English in addition to the tagging system. Best practice is to use a mix of both. You can now use natural English as tags e.g "holding black pistol in left hand", "beautiful face with sharp jawline".
In-Game Images: You can also use mana to illustrate in-game images. The model works both for SFW and NSFW images. Currently only supports 1 character at a time.
Mana has also been added to Mana studio for local character creation.
New Text Models and other changes
-New Texts models added: Gemini 3.8 Flash, Opus 5.5 Sonnet 5.5, Mimo 2.6 Pro
-"Gallery" button added in right side of gameplay screen where you can now view/download last 20 generated images/videos for your story.
-New character sprite surfacing in-game now use reinforced learning to reward characters that are picked more often for the specific role.
-Deleting a character that is used in public stories now requires you to pick a replacement character.
-NPC status calls now updated to only charge 0.1 coin fixed cost per turn instead of the usually 1-4 coins. NPC status model can no longer be changed.
Vector Memory Update
Ever wondered which details or lorebook entry from your story the AI will remember next?
You can now simulate an turn when playing a game and preview which lorebook entries and past story memories would be included, shows their match scores, and helps you see why other memories were left out.
The Recall Simulator previews which lorebook entries (all users) and past story vector memories (tier 2+) would be included, shows their match scores, and helps you see why other memories were left out.
Try a line like “Ask Elara where she got her necklace,” then explore the results: - See selected memories and near misses. - Preview the recalled information that would be sent to the narrator. - Adjust how closely memories must match your input. - Choose a similarity threshold or a set number of top matches for lorebook searches.
Nothing is added to your story when you simulate. Experiment with the controls, then choose “Save to my account” when you’re happy. Saved recall settings apply across your account.
Find it in gameplay settings: Memory Settings → Long Term Memory → Memory Recall → Open simulator.
Custom Backgrounds
Stock background images no longer! You can now create custom backgrounds for your stories or compete for free coin rewards by filling locations on the Background Bounty Board.
New Text Models
2 New text models for paid users! (under "gated" models)
-GLM 5.3 Uncensored: Abliterated version of GLM 5.3 with safety refusals minimized. Tends to reason more. Slightly more expensive than base GLM 5.3.
-Muse spark 1.3 Contributor: Heavily subsidized version of muse spark 1.3 (GLM 5.x level model) allowing meta to train off your messages. So cheap it's below flash model pricing.
New Game Mode: Agent Mode
One story. A whole cast with minds of their own.
Meet Agent Mode, a new way to roleplay where each character is their own AI. They have private memories, and personal goals. They decide what to do and say, then a narrator brings their choices together into a single turn.
-Private perspectives: memories are built around what each character sees, hears, or learns.
-Independent goals: characters pursue their own interests across turns, not just react to your latest message.
-Your choice of cast: assign different AI models to different characters.
-Peek behind the scenes: Use the edit story menu to view and edit each character's memories, summaries, and goals.
Imagine a tense tavern conversation: one companion knows about a betrayal, another suspects nothing, and the person responsible is trying to keep their cover. Same scene. Different perspectives.
Available to subscribers. Agent Mode runs several AI calls per turn, so coin costs can be higher. Turns with multiple acting NPCs receive an automatic discount credit afterward.
Character creation update: Alterative outfits
Armor for a dangerous journey. Formalwear for a royal ball. Something comfortable for a quiet night at the inn.
Your characters can now have a wardrobe.
Alternative Outfits let you create multiple looks for the same character, without making a separate character for every outfit. Each look has its own poses and expressions.
Write when a character should wear an outfit and when they should switch back. Mana's AI uses your instructions to change their on-screen look as the story develops.
New Feature: Detailed Story Tags
Finding the right story should be much easier now. Updated search page.
We added a new tagging system for public stories, so stories can be discovered by genre (e.g Fantasy, Isekai, Overpowered MC) story themes (e.g netorare, time stop, battle), parody/universe (e.g Demon Slayer: Kimetsu no Yaiba, pokemon) and character tags of starting NPC casts (e.g blonde hair, gyaru, armor). Search can combine tags too, which means you can narrow things down by the exact vibe they are looking for.
By the time you see this post, we will have used opus 5 to classify and assign tags to existing public stories as a starting point.
Story creators: please feel free to review your templates and change the tags if something does not feel right. AI tagging is useful, but you know your story best. The easiest way to review your tags is to open your template in Mimi > Finalize tab. Changing tags do not incur AI moderation check.
Please note that nsfw 18+ tags may be hidden depending on your safety filter in profile settings.
QoL changes 8/17
-XP bar to next skill point now shows in the hardcore mode skill point pop up.
-You can now save custom expression for character creation as presets to reuse.
-Generated custom expression images no longer deletes on changing expression name.
-Story creators can now see the total number of messages of a specific story in templates page.
-Story moderation changed slightly, less story will be classified as "explicit 18+" and be correctly classified as "suggestive".
-Mana studio Windows desktop app now open to all. (create character images for free using your own GPU). Requires users to have at least 2 completed characters before allowing access.
New Text Model: DeepSeek V4 Pro 0813
New version of Deepseek v4 pro. Sensitive to reasoning effort.
Medium = 3-5 seconds of planning. High = 10-30 seconds of planning.
Medium effort is set as default. Will replace deepseek v4 pro free if players find it better.
Story Creators: Story Preview Page Update
Click here to see it in action
You can now use custom images for story description and character descriptions. Tell your world through images.
Upload images for story guide pages for maps, lore, factions, timelines, and anything else players should see before they begin. You can also add visual dossier pages for each character, replacing written character details on the preview screen while keeping the original text available to the narration AI.
Use your existing chatgpt subscriptions to help generate the custom images.
Need help creating them? The story editor can copy ready-made template instructions and download your character reference images for use with ChatGPT or another image tool.
Open one of your stories and choose Edit to try Image Description Prompt and Image Previews.
QoL changes & Bug fixes
-Auto character surfacing is enabled for hardcore mode. Please report any bugs.
-You can now choose specific provider for certain models. It affects the quality (slightly), speed and cost of the model. Currently rge option is enabled for only GLM 5.2 and Deepseek v4 pro.
-Mimi created and auto surfaced characters in stories are now slightly more detailed.
-For tier 2+ users: rewinding/redoing/deleting a turn now erase the vector memory generated for that turn.
New Game Mode: Hardcore Mode
When you attempt something risky in Hardcore mode, the story may pause for a cinematic 2d20 skill check.
Every character now tracks three evolving relationships with you: Trust & Respect, Fear, and Attraction. Your actions and dice outcomes change these relationships, influence how characters respond, and can affect future rolls.
Hardcore turns are permanent:
-Settled dice results cannot be rerolled with Redo or by refreshing. Story context is read-only.
-Previous turns cannot be rewound or deleted.
-The cast is fixed once the story begins.
Risky rolls earn XP toward additional skill points.
Most visual stories let you choose Hardcore before starting unless their creator has marked them Realism Only.
Comments System
Public stories now have their own community discussions.
Story creators are identified with an Author badge, and replies remain organized in easy-to-follow threads. Comments are automatically moderated to help keep discussions welcoming.
The new Inbox icon at the top of Mana will notify you when someone comments on your published story, replies to you, or continues a thread you joined. Open a notification to jump directly back into the discussion.
Free Coins: Reddit Post
Share your experience with Mana and earn a one-time 2,000 coin reward.
Open Free Coins → Reddit Review, write a review in a subreddit related to AI roleplaying, then submit the link to your post. Please follow the subreddit’s rules and leave the post up while it is reviewed.
After 24 hours, we’ll verify that the post is still available. Once approved, the coin reward shown on the page will be added directly to your balance.
You’re also welcome to include your Mana referral link in the post. This reward can be claimed once per account.
Free Coins: In-game Image Thumbs Up
We need images as training data to our train own own image model. See an good Stella image? Give it a Thumbs Up!
Eligible generated images now have a circular Thumbs Up button beside Download and Close.
Images generated with Stella using Mana Default, 2D Real, or 2D Real Dark can qualify. Both single-character and two-character images are supported, including SFW and NSFW generations.
Every eligible image you submit earns 10 coins.
Training submissions are anonymized, and usernames are not collected. We only log the image and tags used.
Each generated image can only be submitted once. Please only like images you genuinely consider good—using the feature solely to farm coins may result in access being removed.
Tier 2+ Vector Memory Update
Tier 2 and Tier 3 members can now see the exact vector memories created and used.
Open Memory Settings → Long Term Memory → Vector Memory to browse the facts Mana has automatically extracted while you play. Memories are organized into turn groups, with the newest groups shown first.
Memories used during your most recent narration turn are highlighted, making it easy to understand what information the AI recalled. You can also edit a memory or permanently delete anything you no longer want retained. Edited memories are reprocessed so future semantic searches use the updated information.
Rewinding or deleting gameplay turns now removes memories created from those erased turns as well. Replacement turns can then be learned normally, preventing outdated “ghost” memories from affecting the story.
Please report any bugs with this feature or if you believe your vector results are of bad quality.
New Text Model
Added Deepseek v4 Flash 0731; the next version of deepseek v4 flash. Added as paid and free model.
Planning/reasoning duration before output is VERY sensitive to "reasoning effort" compared to previous version.
Bounty Board
Mana’s stories are always looking for the right cast. Now you can see which character roles are in demand—and get rewarded for filling them.
Open side menu→ Bounty Board to browse casting gaps discovered during real gameplay. Sort by Most Requested or Weakest Match, then review the requested era, gender, species, demand, and current claim count.
Already have a fitting character? Make sure their classified role exactly matches the bounty, then submit them for review. You can also create a new game-ready character for an open role.
Missing Male character roles are currently incentive. Every bounty has only three claim slots, and each player can claim it once.
Custom Character Expressions
Each Characters now supports up to 8 custom expressions in addition to the default expressions.
Open My Characters, choose a character, and select "Manage" under Custom Expressions & Outfit.
The same panel also lets you fine-tune default-expression tags and adjust sprite zoom. After adding a custom expression, generate its missing sprite assets to return the character to game-ready status.
New Text Model: Kimi K3
#1 ranked open source model on RP/Creative writing benchmarks. "xhigh" reasoning effort seems to output response instantly while "max" reasoning effort seems to spend 5-10 seconds planning before writing.
New: Two Pass Mode
Get stronger story writing and more reliable scene visuals with Two Pass Mode.
Each visual-story turn is split into two focused steps: your storyteller writes the narration first, then a Scene Formatter adds character sprites, expressions, and backgrounds.
Enable it in Gameplay Settings → Settings Panel → System Settings. Models marked “2P Mode Only” use this workflow automatically.
Two Pass Mode takes a little longer and uses additional coins for the formatter call. Leave it off whenever you prefer the faster, lower-cost Single Pass experience.
4 New Texts Models
Doubao Seed Character: Lightweight model created by ByteDance specifically for RP. Creative in dialogue and tension. Struggles with logic and multi character planning.
Aion 3: RP fine tune of GLM 5.2. Very expensive compared to base model.
Aion 3-mini: RP fine tune of deepseek v4
Opus 5: Strongest model in terms of EQ. Nerfed nsfw writing ability compared to 4.6
To help manage the ever growing collection of text models, we added the ability to favorite models.
In-story image gen: creativity slider
Stella image model's single character generations now gives you direct control over how closely each image follows your scene.
Open "See", choose Stella, and select one character. The new Creativity slider appears after Angle:
• Move left for more creative variation. • Move right for stronger context and prompt adherence.
Your choice is remembered separately for each art style.
New action type: Speak
Speak with any character in your story real time while they maintain their voice, personality and memories.
When the session ends, the conversation is added to your story history so the next narration turn can continue from it.
Characters need an assigned voice to use Speak. You can assign or change one under Gameplay Settings → Character Management.
You can configure settings for this new feature under Gameplay Settings → Speech.
Billed per exchange, but costs are actually lower than a text turn with the same AI model.
Free User Summarization Is Here
Now long-running stories can now maintain their continuity on the free plan.
When the AI’s working memory fills, Mana will prompt you to Compress Context. This folds older story turns into a compact memory while keeping recent scenes intact, giving your story room to continue.
You can review and edit the generated memory before saving it. Once saved, it will be included in future narration turns and can be edited later under Gameplay Settings → Memory → Summary.
If your selected summarizer is a free model, it will use the Free Queue and cost no coins. Paid summarizer models use coins normally.
Upgrades to paid user's memory stack is on the list next!
Meet Mimi, Your AI Story Creator
Creating a full story is now a conversation. Tell Mimi your idea and she can draft the title, description, starting prompt, player info, characters with sample dialogue, lorebook, and backgrounds, and gameplay mode—then you can review or edit every section before publishing/playing.
Mimi can also load one of your existing templates for a rewrite!
Custom Story Thumbnails Are Here
Your public story cards can now look exactly the way you want. Now supports uploading custom thumbnails.
Bring it to life by animating your thumbnails with Grok Imagine or Seedance. Animated thumbnails play as seamless, muted back-and-forth loops on story homepage.
New Model: Grok 4.5
New text model added Grok 4.5 - 25% discount on this model for 5 days. Give it a try!
New Feature: Black Market
Uses alternative routing to offer ~70% lower coin costs on select Claude and Gemini models. The trade-off: they may be slower, less consistent or have stricter safety filters than the regular Credit routing.
New Feature: Auto Add Characters
Stories can now introduce new characters visually during gameplay. When someone new appears, Mana can cast a matching character from the character library and add them to your story via image vector search.
Want a different look? Tap the small character edit badge on the sprite to browse alternate visuals. Characters with voices show a mic badge, and your choice locks in when you send your next turn.
You can turn this on or off anytime in Gameplay > System Settings > Auto Add Characters.