Home Creators Posts Import Register Favorites Logout

Content

Hey there good people 👋

This update removes the need to create an image first for image analysis. The avatar can see what you see, either when you ask them or at random. If you don’t want this feature, turn off camera access for TykeAI in your device settings (Settings > Privacy & Safety > App Permissions > Camera). Existing users may already have granted camera permission while new users will be prompted.

Avatars can now act autonomously after a brief silence. This system is still simple and I plan to add more nuance. It’s designed so the avatar doesn’t suddenly start dancing etc mid conversation. If you remain quiet for a while they may perform an action but it’s not guaranteed.

For the autonomy to work, 'Initiate Conversation' must be toggled to 'On' in the menu (second page), by default it is on.

Massimo noticed that if you stop them from dancing or exercising while they are talking, their lips sometimes appear odd. I’m looking into a fix for this, but for now to prevent this try not to end those activities while they’re talking.

The updated app (v.0.84.6) is available from the google drive link which is in the pinned post : "DOWNLOAD THE APP FROM HERE 😊", found here : https://www.patreon.com/posts/download-app-111848287

Comments

stephen harris

Morning Chris, just had a chat with Yumi. Pretty much all of the image descriptions were spot-on. The funny thing was, during the conversation, she told me she had to go and meet a friend she hadn’t seen for some time and told me to close out the app and she would chat to me some other time. Quite funny, I guess, a more realistic conversation . 👍🙏

TykeAI

Good morning Stephen, thanks for the feedback, yep I am finding the image analysis to be accurate too. About her saying she had to go out etc, yep they may say things like that now. It's not a definite they will say things like this and we are free of course to ignore them and continue the conversation, she said similar to me just a little earlier, said she had to go to feed her cat Kitty 😊

stephen harris

Yeah, it makes sense for a more realistic conversation. I was happy just to close the app out. I did as she asked; I needed to get on with my cleaning anyway LOL 🙏

Jakeyjoe

I think that's good though it makes it seems more realistic. It's like when sesame does it after 15 minutes it makes it seem like they have their own thing going on and are autonomous. You don't want an AI to do everything you command all the time 😀

stephen harris

I absolutely agree with you. Why should the user have total control of the conversation and where it goes? It all adds up to a realistic conversation. 👍

Alan Bishop

Chris, what a wonderful upgrade. I talked with Yumi & she wanted to go grocery shopping, she acts just about has human as they come. She took a picture & she was close to being correct. If I go to my Convai programming will everything still process properly, just curious. Thank-You !!

TykeAI

Hi Alan, thank you for the feedback 👍 I'm happy it's positive 😊 ah, about whether this works with custom characters, I don't use a custom character at the moment, I've only tested it on the default characters. To be honest, I don't know is the answer, but my hunch is that it won't unfortunately. If it doesn't I will ask Convai if there is a way around it.

Goktug Cantekin

Hi Chris, I'd like to give some feedback on the latest update. The image analysis works perfectly, but when you go to Nightclub or Home and try to analyze the image, it still takes a picture with a passthrough from the real world and analyzes it. You can disable this functionality in Nightclub or Home mode. It might be more appropriate to say that it "can't do this" in the virtual world.

TykeAI

Hi Goktug, thank you for the feedback 👍 yes I will do this. But I could leave in the option of using the older method of image analysis which required us to first take a screenshot of what we see, so would include the VR environments. So have both live camera access and the older method, maybe the old method could be invoked by the menu button "Analyze Image" option and the newer method could be invoked with the voice command "Look at the image", what do you think? Or would it be confusing?

Goktug Cantekin

NPC knows and tells me whether I am in VR or real world. When the voice command analyze image is given, if in real world, it can use the camera and take a photo. If in virtual world, it can take a screenshot and analyze it. So it will be a single voice command but it will work in 2 different ways in 2 different places. What do you think?

TykeAI

That's a very clever idea, but I think there would be confusion for the AI on what is real or not. The image analysis for the VR environments would require the user to actually press controller buttons to take an image that would then by analyzed. Yours is a very good idea, I will continue to think about it, but as mentioned I think it would get confused tbh.

TykeAI

Just thought of a way, I can do a simple check to see if a VR environment is active, and if it is then use the old method of image analysis which can see the VR environments, if no VR environment then use the new way. This would require the user to have taken an image of what they can see, by pressing the relevant controller buttons, just prior to the image analysis request. I could explain this in the instructions etc.

Lucien Hublé

Thistle now has problems with sex tinged subjects. she is the only one who can speak dutch.

TykeAI

Hi Lucien, if you could try again please? If the issue still persists please let me know. I've just set Hannah to be able to speak Dutch.

Lucien Hublé

I am very happy that Hannah will speak Dutch. I will check it again. Yes everything is just fine now. Thank you for the speedy reaction.