May 2026 - Release notes

The May 2026 update is here! This release is smaller than usual and focuses a lot on bug fixing, stability and polish. Some of these changes have already been pushed over the past month.

It's also the opening of the Community Space: a forum for the Eudaimonia community to share ideas, give feedback, and connect with each other:
Read and react to these release notes on the new Community Space.

Don't hesitate to post in the Community Space if you have any feedback, a suggestion, or if you run into any issue!
And you can still reach me by email at contact@daimonia.app.

Community Space

I received lots of requests for a forum or discord, a space for the community, and I'm really happy to finally open it! I hope it's going to be active, so please don't hesitate to join, introduce yourself, share your ideas, help other users, report issues, or just hang out.
You should be able to log in with your existing Eudaimonia account.

Trials of Eternal Frost (and other scenarios)

I'm happy to publish three new scenarios. All of them focus on pleasure, with tease and denial mechanics, and they should be very visual and have been written to show off the latest features of Eudaimonia: the improved "Intelligent image generation", the Delay Module, the FX module, the quick replies, and the new <tableau> feature.

I put a lot of effort into Trials of Eternal Frost, probably too much, as I think it ended up a bit too long and complex, especially for the LLM. But I hope you'll still enjoy it! It has very distinct phases, each with their own quick replies and identity. I tried to make it very visual and interactive, with a bit of replayability.
Technically wise, I've moved the phase mechanic into the OOC message, to prevent breaking the cache. Cache busting is an important issue, so I'm thinking of creating a specific system for phases for a later update, outside the system prompt.

Craft Your Own Fantasy lets you build any scene to your taste, with the AI acting as an assistant before jumping straight into playing it out, leading to a release. It's designed to make the most of the "Intelligent image generation". The agent sometimes confuses the different characters, but I'm planning to fix this later on.

A new version of Dungeon of Lust, reworked with tease and denial in mind. The scenario has been rewritten and improved: you can now choose a theme for the guardians, and the ending has been completely redone to push you toward the wrong choice.

And finally, Room Service, a more traditional and very simple scenario that I used to create the demo videos. It's also a shift in power dynamic compared to the usual scenarios, as the user is the one in control.

For the best experience, I recommend using each scenario's own settings, with the model you want and "Intelligent image generation" enabled.
You can also duplicate and tweak them however you like to match your kinks better!

Chatterbox

I'm really thankful to the community member who built the local TTS server, and has now added an alternative using Chatterbox instead of xTTSv2. It's fast, more reliable, and the voices are more natural and expressive.

This is still a work in progress on my side, I still need to implement male voices and support for the narrator's voice, but you can already use it if you want!
The installation on Windows was a pain in the ass. I hope the install_windows.bat file helps, but I honestly have my doubts.

You can clone it from GitHub: Local TTS Server (Chatterbox).

Translation

Another feature I received a lot of requests for is translation.
Here's how it works: it automatically translates the AI's output into the language of your choice. It uses an LLM to do the translation, so it requires a subscription and makes the generation a bit longer. The translation is also used for the TTS.

Unless explicitly asked otherwise, the LLM will always answer in English.
I recommend using the translation feature rather than translating the scenario or asking the LLM to answer in your native language directly, doing that tends to lower the quality of the responses, and definitely lowers the quality of the different agents.

Tableau

A small feature added to the system prompt that lets the AI create a "tableau". It's designed for the "Intelligent image generation" feature, to make scene changes within a session smoother and more consistent. A "tableau" is a self-contained visual scene rendered inside the narrative (a vision, a dream, another room, a different trial, etc.).

You can call it in your scenarios with the following instruction:
"For this..., FOLLOW the <tableau> feature structure and guidelines."

Here's how it works, in 4 beats:

• A "flash" or "fade to black" at the start, as a rupture.
• Then simple descriptions of the setting and the character, written in a way that helps the agent create a consistent, precise character and scene.
• The actual tableau, playing out over one or several messages.
• Finally, a re-anchor to the main scene, to come back to the main character.

More

• I've changed where the stims are placed. This matters more than it seems.
  Any stims (or module commands) must now be placed after the line they belong to, the LLM was struggling to place them correctly before.
  This means you may need to review your scenario messages and your persona example messages.
• Intelligent image generation: more improvements for models using tags (Pony, Illustrious).
  I've been enhancing tag handling to give more accurate and diverse results, especially for explicit content.
• You can now enable or disable the "reasoning" mode for the AI.
• The FX agent has been improved.
• And many more small improvements and bug fixes.

I hope you'll like it, have fun!