0.34.0

Visual Agent Refactor

Visual agent refactor adds image editing with reference images and image analysis capabilities. Image editing allows modifying existing images using reference images. Image analysis extracts information from images for use in generation.

Supported backends: ComfyUI, Automatic1111, SD.Next, OpenAI, Google, OpenRouter (with possibly more added still)

Visual prompt instruction can now be customized through new Visual Style templates in the Templates editor.

Nodes for image image generation have been added to the node editor.

Visual Library

Replaces the current image queue and allows for viewing, (re)generation and iteration of images.

Adds scene visual asset management, with tags, categories, image analysis and preparation of reference images (for the AI to pick and use during generation when an image editing backend is availabile)

Overhauled character card imports

Character card import system completely refactored. Analyzes card to detect multiple characters present in the card, allowing selective import of detected characters.

• Support import of alternate greetings
• Support import for character books (lore books)
• Improved character generation now uses card description, greeting texts, and character book entries together to determine character descriptions, attributes, and dialogue examples.
• Improved player character selections - can be set using the default template, selected from detected characters in the card, or imported from another scene.

Notes

• The Templates editor has been moved out of the World Editor and is now available from the main navigation and no longer requires a scene to be loaded.
• Runpod integration has been removed.
• Something in the pipeline for TTS generation (chatterbox and f5-tts) changed where there now is a dependency on torchcodec and ffmpeg. Install.bat for windows will take care of this automatically, linux users will need to ensure ffmpeg is installed on their system if they want to use either chatterbox or f5-tts for audio generation.

Bufixes and other improvements

• scene assets are now managed in a separate file that all scenes in the same project share (assets/library.json)
• fix issue where narrator would use wrong preset / system message
• prompt for unified api key configuration during client setup
• Improved token/s calculation
• add Reset shortcut to Save menu
• add save required indicator to save menu
• fixed "'_SceneRef' object has no attribute '_changelog'" error
• client selection in agent config will now only show enabled clients
• app now properly locks inputs when client configuration is missing
• improvements to director chat
• shared world: button to share/unshare all characters
• shared world: button to share/unshare all world entries
• shared world: added episodes which is basically just talemate's version of alternate introductions
• narrator: story progress now respects the generation length setting in the narrator agent config
• character state reinforcements: added require_active setting that requires the character to be active for the state to be reprocessed. This defaults to true.
• Added Reason prefill setting to client config to coerce reasoning responses
• fixes #222
• fixes #223
• fixes #175
• fixes #234
• fixes #129 (maintain a scene with characters to import from if you want to manage multiple default player characters)
• fixes #232

Node editor

• added node modules for image generation
• added node modules for scene asset management
• pressing w while holding a wire will now create a watch node connected to that wire
• pressing s while holding a wire will now create a set state node connected to that wire

0.33.0

Director Chat

A chat interface for conversing with the Director agent about the current scene. The Director can execute actions through 25+ specialized node modules covering scene queries, character/world state updates, game state management, narrative direction, and history modifications.

Minimum recommended parameters: 12k+ context, 32B+ model with reasoning enabled. Ideally 100B+ models for best results.

Accessible through the director console.

Scene Changelog and Restoration

Tracks all scene changes over time using delta compression. Stores incremental changes between revisions in segmented changelog files. This allows scenes to be reconstructed fully to a specific point in history and will be used for restoring scenes as well as true forking of scenes.

Shared World Context

Allows marking characters, world entries, and static history entries as "shared" to synchronize them across multiple scenes in the same project.

Shared elements are exported to a dedicated context file that other scenes can reference. For characters, supports granular sharing at the attribute and detail level.

Shared context files can be created and managed in World Editor -> Scene -> Shared Context.

Other imrpovements

• Agent persona (only for director currently)
• Pin decay - pin will stay active for N turns without having to be re-evaluated. Once decay time has run out - re-evaulate pin condition or drop the pin.
• Improvements to data structure handling in LLM responses
• Game state variable editor
• ContextID system for unified management of context (backend only, currently not exposed to the UX - director chat uses it extensively)
• Pressing ctrl+up arrow/down arrow wil lcycle through previous messages in the scene chat input
• Added summarizer / editor / director message access to scene toolbar. This is useful for reviewing revision changes, scene analysis and director guidance.
• Scene forking function will now use the new changelog system to reconstruct the scene to the specified revision. For pre 0.33.0 saved scenes this will still do the old shallow fork.
• World Editor -> Context DB has been made read only (other than pin management). Context must be updated through the appropriate tools in the World Editor.
• Memory agent semantic retrieval improvements. Will now compare phrases as well as the whole messages and can be configured to compare more than just the most recent message.
• improvements to autocomplete when prefill isn't available

Bugfixes

• Fixed issue where drag and drop scene / character cover images would no longer work
• some layout issues in world editor
• Enforce Context DB clean up on scene load
• Openrouter api key going from unset to set will immediately fetch models
• Fixed remaining "World state manager" references in the UX to "World Editor"
• Scenes that no longer exist are automatically removed from recent scenes list
• Contextual generate will default to the scene writing style if its set.
• Fix ai function argument conversation that would cast everything to string.
• fixes #231 - looping error when attempting to delete some clients
• fixes #230 - client can now lock to a specific template

Node editor (improvements and bug fixes)

• 70+ new nodes across 12 categories (Context ID, Scene/Character access, AI functions, data structures, etc.)
• Module library overhaul to treeview display in sidebar
• Show when required inputs are not connected (red links)
• Certain nodes can now be ALT+SHIFT+Dragged to spawn a counterpart of the same node. (e.g., Set State to Get State and vice versa)
• Collector nodes for easier collection of values into lists or dicts without needing to nest nodes.
• Clicking outside of node property editor will no longer discard changes automatically.
• Fix issue where graph state wasn't displayed correctly inside inline functions.
• Errors in event nodes should no longer be allowed to cause infinite loops of failures.
• Deleting a module while its loaded will no longer result in an error.
• fix issue where editing graph properties would cause the graph to reload and lose unsaved changes
• Fixed issue where argument nodes would not convert to their type correctly.
• Switching from node editor to world editor will no longer cause the changes to the graph to be lost.
• ALT click on group title to snap to nearest other group will no longer cause overlap with other groups.
• Fixed JSON error message not resetting in node property editor.
• Node search improvements
• Get node can now properly access tuple and set items.
• Auto resize new nodes to fit their title.
• Auto resize new nodes to fit all their inputs.
• Node property choice fields are now sorted alphabetically.
• Node property field will no longer show its original value if the matching input is connected.
• Added group presets
• Contextual Generate node will now error nicely when the required context_name is not set.
• Fix issue where Ctrl+Enter to submit in text editor nodes would cause extra new lines to be added.

0.32.3

fixes #218 - Right click does not working in node editor canvas (or documentation out of date)

0.32.2

fixes #217 - koboldcpp connection error

0.32.1

Bugfix release

• fixes #212 - LMStudio connection issues
• fixes #211 - installation issues when path contained spaces or other special characters on windows
• fixes #214 (thanks @IcemanOakenbear)
• fixes issues with character creation
• some tweaks to scene analysis and director guidance prompts for conversation
• adds GLM 4.5 templates

0.32.0

TTS Agent refactor

The TTS (text-to-speech) agent has undergone a complete refactor, adding support to additional apis, per character voice assignment and speaker separation. Voices can be managed through the new voice library.

Supported APIs:

Local

• F5-TTS (zero shot voice cloning from .wav)
• Chatterbox (zero shot voice cloning from .wav)
• Kokoro (predefined voice models that can be mixed to make new voices)

Remote

• Elevenlabs
• Google Gemini-TTS
• OpenAI

The director agent is also now able to automatically assign voices to new characters, with selection based on tags managed through the voice library.

I made the decision to drop XTTS2 as it seems to have been officially discontinued AND both F5-TTS and Chatterbox do the same thing, but with vastly better quality.

Note on RVC

RVC support was investigated, but has NOT been added at this point. It currently depends on some outdated libraries that clash with what we require. Support for remote APIs could be added, but their setup proved to be fairly finnicky, and with the way zero shot voice cloning is going, it may no longer be necessary, so i am deciding to wait and see. Feel free to reach out in discord if you feel strongly about RVC support.

• fixes #170
• fixes #182

Support for reasoning / thinking models.

Support has been added for reasoning / thinking models. This has been done across all client types, but some clients may need some extra configuration for this to work correctly and may still lead to wonky behavior. Generally the experience should be good though.

There is a new Reasoning tab in the client configuration UI. Reasoning NEEDS to be explicitly activated through there if you have a reasoning model targeted that you want to use.

Scene export / import changes

Scenes can now be exported as complete packages, containing all nodes, assets and info files. Likewise such packages can also be imported through the load via file upload option on the home view. For backwards compatibility standalone JSON scene files can still be exported and imported also.

Underlying config handling refactor

Code housekeeping change that should be mostly invisible to users, but should lead to improved stability and less effort when extending configuration options in the future.

Smaller features / Bugfixes

• Disallow pointing multiple talemate clients at the same backend (e.g., multiple devices / browsers at the same time). Talemate is currently not designed to handle this, and people have ran into save file corruption and other issues because of it. So for the time being it will be enforced that only one client can be connected. This does not mean this is the final state of this, but necessary while details and use cases are decided.
• Simulation suite: fix issue where characters were added as inactive
• OpenRouter: fix issue that would lead to decreased quality in generation and data structure response handling
• OpenRouter: Allow to specify providers
• Fix jinja2 error that could pop up during auto direction generation
• KoboldCpp client now has its own default template according to their KoboldAI format that will be used as a default if the appropriate LLM template can not be determined automatically.
• Agents that can be disabled / enabled can now be ctrl clicked in the agent list to toggle them on or off.
• Visual agent - fix issue of scene cover image being replaced automatically with each generation
• disable scene analysis / guidance when building image gen prompts
• migration from vue-cli to vite (thanks @pax-co)
• example dialogue editor input changed to be more obvious about support multiline input
• fixed issue with character generator where the "Player controlled" flag was not respected
• fixed issues with scene analysis cache
• fixed issue where editing a message could cause parts of the message to be lost
• docker package release workflow (thanks @luisbrandao)
• openai GPT-5 models added

Node-editor

• As String node added for string conversion
• Generate TTS node added for text-to-speech generation
• Get Narrator Voice node added
• Get Voice node added
• TTS Agent settings node added
• Unpack voice node added

Discord Community

Need help? Join the new Discord community

0.31.0

Installable node modules

Added very rudimentary way to register node modules as packages so they can be installed into a scene.

Added mods tab that becomes available once a scene is loaded and allows installing / uninstalling of such modules. The Dynamic Story node module has had a package added to that effect and you should no longer need to mess with the node editor to set it up in a scene.

Improved install process on windows

Running start.bat will now automatically run the installation if dependencies aren't installed, making it the main entry point start and install talemate.

The install.bat and update.bat scripts have been restructured to use embedded python and nodejs, pulling the correct versions, eliminating requirements for build tool installation or maintaining multiple python / node.js versions at a system level.

Failures at any point in the installation process should now be more obvious and stop the process, also.

History Management

Can now add, edit, remove and regenerate entries in the history tab of the world state manager.

Additionally entries based on summarization can be inspected to show their source messages.

General

• fixes #166
• fixes #189
• fixes #176
• fixes #194
• fixes #195
• fixes #186
• fixes #185
• fixes #192
• corrupt world state templates should no longer break the world editor
• added toggle to disable automatically adding the default character to blank scenes
• narrator prompt improvements
• conversation prompt improvements
• scene analysis prompt improvements
• RAG improvements
• continued template file cleanup
• fix issue where example dialogue generation in the actor tab was no longer working
• LLMStudio - fix some prompt template issues
• Anthropic - enabled coercion / prefills
• Google - enabled coercion / prefills
• Google - support API key auth (google cloud auth is still supported)
• allow import of chara_card_v3 and attempt to future proof
• improvements to summarization
• improvements to editor unslop revisions
• fix issue where "make character real" from world state listing would no longer work
• any contextual generate instruction dialog should now let you specify generation length
• Dependency management was switched from poetry to the much faster uv. (Thanks @ztripez)

Node Editor

• fixes issue where errors inside custom node graphs could hang talemate
• fixes issues with a bunch of math nodes that would cause them to run even though the wires to them were inactive
• fix issue with DynamicInstruction node that would cause it to run even though the wires to it were inactive
• fix issue where editor revision events were missing the template_vars value
• jinja2 templates existing in templates/modules can now be properly loaded
• Emit System Message node - allows for communication of messages to the user outside of the context history

Discord Community

Need help? Join the new Discord community

0.30.1

Small bug-fix release to address some installation issues

Bug fixes

• fixes cyclic import issues in some environments
• fixes some build issues related to old numpy version

0.30.0

Node Editor

The backend was refactored to a node based architecture, allowing for more complex and dynamic scenes and customizable / reusable modules.

This is the first iteration of the node editor and a lot of kinks still have to be worked out. The node editor is accessible from the creative mode once a scene is loaded.

Note: There is no good way to export / import node modules yet, but this will be added in a future version.

Revisions

Revision action added to the Editor agent. This action, when toggled on, will analyze text for repetition or unwanted prose and revise it accordingly. Unwanted prose is defined through the writing style template assigned in the scene settings.

Auto-Direction

The Director agent can now automatically direct the scene based on the current state and intention of the scene.

This is experimental and a work in progress and my personal testing with it so far has had mixed results.

The goal is to test the waters towards giving the reigns to the Director agent to direct the scene as it sees fit.

Other improvements / fixes

• Inference preset groups
• AI function calling improvements
• Clients can now configure data communication to be in YAML or JSON format
• Simulation Suite V2 (v1 still exists) - remade using the new node editor and the new AI function calling.
• Infinity Quest Dynamic Story V2 (v1 still exists) - remade using the new node editor and a re-usable node for story randomization that can be plugged into any scene.
• Director guidance cache
• fixes #179
• fixes #183
• fixes #173
• fixes #172
• fixes #165
• fixes #160
• fixes #178 - generation length can now be configured through the creator agent settings - also generation should have been improved in general
• fixes #187

Discord Community

Need help? Join the new Discord community

0.29.0

• fixes #163
• fixes #156
• fixes #157 - removed extra padding, seemed unnecessary - configurable is still on the table if feedback demands it
• fixes #151
• fixes #152
• fixes #162
• fixes #168 - intro text should be changed through the world editor
• fixes #171 - added button to request choices, and whatever character is active will be target of the choice creation
• fix bug where canceling a generation would auto cancel the next attempt as well
• scene analysis to help guide conversation and narration (summarizer agent)
• context investigation rework (summarizer agent)
• narrator improvements (prompts revisited, context polished)
• narrator long term memory options exposed (similar to conversation agent)
• narrator can now use a writing style template (set through the scene settings and a on/off toggle in the narrator agent)
• director can now guide actors based on scene analysis
• director can now guide narrator actions based on scene analysis
• director long term memory options exposed (similar to conversation agent)
• summarization with analysis to improve quality of layered history summaries (toggleable)
• summarizer long term memory options exposed (similar to conversation agent)
• long term memory caching
• adds favicon
• favicon will change while talemate is busy
• improvements to context investigation
• implement backend to separate function orchestration from creative writing
• add suggestions tab to world editor - agents can now propose updates to character attributes and description based on story progression.
• add DeepSeek API client
• LLM system prompts can now be overwritten in the global app config as well as specifically within each client config.