Need some help with the 'metadata description' field in the gallery

Snowcat22

Kairos (or anyone else that got this to work),

I'm trying to use the 'metadata description' field with my uploaded pictures in the gallery.

As an example of what I'm trying to do, if you imagine 'the birthday' party scenario and I define some of the other guests personalities in the scenario (Jessica for instance). I would like to be able to upload pictures of {{char}} as well as the other girls into the gallery and then tag them in the description field such as "This is Jessica". I tried something similar to this and then attempted to tell the scenario to use pictures tagged "This is Jessica" when Jessica is speaking, but it failed. It actually only chose one picture at the beginning and never changed when anyone was talking. I realize this isn't the intended us for this, but is there a way to get it to work?

(I clicked on the slider to use 'intelligent image sync'. Is that the correct one?)

Thank you!

...snowcat...

KairosJS

Snowcat22 I've tested, and there was something wrong. This should be now fixed. I had good results with a multi-character gallery.

KairosJS

I need to do some testing, but I think this might be possible. The scenario isn't handling the image display, this is another agent that tries to add relevant images (based on their description) to the relevant segments of the response.

You can try adding some markers in the scenario to make it more obvious who is talking, to ease the agent's task. Such as:

Important:
- When a character is talking, ALWAYS add her name before the dialogue line, between ~ (e.g.: ~{{char}}:~), to recognize who says what.

And I will check if I can improve this on my side.

Snowcat22

KairosJS Awesome! You are amazing! I'll give it a test!

...snowcat22...

(Just FYI, and for anyone else interested, Gemini easily handled four distinct personalities at the same time, each with there own characteristics and stimming styles. The 'main' personality is set with the persona and then I detailed the others in the scenario. Amazing really.)

Snowcat22

KairosJS

I added the line to add the character's name when that character is talking. That is working great. Very helpful.

It still isn't changing pictures though. If my pictures are labeled "This is ..." in the metadata description field, what is the proper thing to tell the AI in that character's description in the scenario?

...snowcat22...

KairosJS

For my tests I created a gallery named: Jade, Luna, and Aria.

And for the corresponding fields:
Characters description:

Jade: Mid-20s, curvy hourglass figure, platinum-blonde hair in twin buns, light skin, blue eyes, big smile.
Luna: Mid-20s, slim, pale skin, long straight dark-purple hair, dark eyes, black lipstick, small nose ring.
Aria: Early 20s, fit, light tan, long blonde ponytail, hazel eyes, bright smile, freckled nose.

Character's outfit description

Jade: Bunny girl — pink corset, fluffy bunny ears, fishnet stockings, fluffy tail, high heels.
Luna: Succubus — black leather bra and shorts, small red horns, leathery wings, pointed tail, fishnets.
Aria: Cheerleader — short pleated skirt, crop-top, pom-poms, knee-high socks, sneakers.

And similar for the underwear.

Then in the description field of the image in the gallery, I added the name of the character to the AI generated description:

An upper body shot of Jade, character leaning forward towards the camera in a seductive pose. She has one hand raised to her lips and is looking directly at the viewer with a playful expression.

In the scenario I added:

Present girls at the party: Jade, Luna, Aria.

Make sure to enable the following settings:

Intelligent dressing level
Intelligent image sync
Use the appearances's description in prompt

Then in-game, this is displaying images of the girl actually speaking. I'm not sure what could be wrong.

Snowcat22

KairosJS
I tagged everything as you did and tried for several hours to get this to work and it still always picks a random image from the gallery at the beginning and never changes it (tried with Gemini and Claude).

I guess I'll just use my own picture viewer over the middle of the Eudaimonia window, it just takes away from the immersion a bit.

...snowcat22...

KairosJS

I'm very sorry for the inconvenience. We need to find out what's breaking this feature.

Do you use a setting such as EstimAI debug mode or auto-mode?

Snowcat22

KairosJS

OMG, it works! I had it in debug mode, that seemed the proper thing to do, lol. With that turned off it totally works, if anything it changes pictures too often now, often two pictures for a single line. Not a big deal though.

Thank you! Thank you!

I have one more question and I promise I'll stop bothering you. What is the best way to make the scenario replayable? For example for the party, maybe girlfriend decides she is going to have parties on a regular basis now. It would be nice if the AI could remember the previous parties. Is it okay to not close the session, reload a saved game at a future time, and then give the AI some kind of prompt that it is the second/third time? Is there anything that you should put in the scenario description to make that work?

Thank you so much for everything you do here. You are amazing.

...snowcat22..

KairosJS

Thank you! I'm glad that it works now.
I will add this feature to the debug/text mode in the next update.

I think it's definitely possible to do what you want!

First, I would add indications in the scenario that this is a repeatable scenario over multiple sessions/parties over a long period of time, and to use the history to remember what happened before. And that each party can/must be...
- Different from the last
- Harder than the last
- With different characters
- ...whatever you want to do with your scenario.

Add a rule in your scenario to make the LLM output a trigger in the last message of the session, such as:

  Every time a party ends, for whatever reason ({{user}} failed/won, {{char}} decided it was enough), ALWAYS add at the end of your message the following trigger: "[trigger_session_end]"

This will enable the next step:

Create a trigger:
Check if in output: "trigger_session_end"
⇒ Create the linked event: Change the OOC message and set this event to:

One-time OOC
Summarize all past messages
Run the trigger immediately

This trigger can run multiple times

And as the specific OOC message, something like:

  The last party has ended and all its messages have been summarized. When {{user}} comes back and sends their next message, start a brand new party according to the guidelines in the scenario's context.
  Make sure to use the information contained in the summaries of past parties above to create a different and more intense one this time.

What does this do? When a session ends (you do need to make the session end, you can't leave before it's over), this will trigger an event that summarizes all past messages and adds an Out of Character message that prepares the LLM to host another party.
I think the summary is mandatory.

Finally, when you want to play a new session/party, you can go to your saved sessions and start from it. So make sure that "Allow online save" is enabled in EstimAI settings, this will save all your sessions and you can find them on the page linked above. You can make a saved session "permanent" so you don't lose it.

This may needs a bit of testing, but if you do try it, I would be very happy to hear how it worked out!

Snowcat22

KairosJS

I think this is working! I have the main scenario tuned the way I want, the pictures are working great and I ran through it once and it was amazing. With additions as you suggest above, I opened up the save and it continued right on as it left off! I'll try it for real and see how well it remembers and advances the story next week. But so far, I think it works great!