Kiskocya
Member
- Dec 2, 2019
- 362
- 839
- 332
- Aug 22, 2019
- 146
- 353
- 157
- Aug 22, 2019
- 146
- 353
- 157
By the way, dj03man, you once mentioned a new game, or rather a sequel to this game. If you're thinking about it, I've come up with an idea to combine this game and the plot of "The Dreamweaver." For example, the main character of "The Dreamweaver" dreams of everything that happens in this game and will travel to the elven kingdom to meet the main character or queen to prevent a goblin invasion or otherwise help with his knowledge of what's about to happen. What do you think of that idea?
- Aug 22, 2019
- 146
- 353
- 157
Hi alsert. Your enthusiams is appreciated!
I really like the concept of the Dreamweaver: one male character, can dive into dreams and shape opinions, story oriented, discover the world and lore.
There's a lot of interest in that genre. So, I'll stay on it and very likely build more stories for this "world" that we're creating.
I don't have a plan yet on what comes next when I finish this game, because there's still quite some work to be done here. What I have learned is that story oriented games really take a lot of time and assets can't be re-used easily.
I'm still learning so we'll see what the future brings, what I can tell you is that I like the dreamweaver concept.
My ultimate goal with the dreamweaver is the "full" freedom. A game where the player decides the interactions himself in pre-generated worlds.
Imagina a VN game (3D is very hard to do with today's tech) where player logs in, choses his character in a fantasy world and has full freedom to chose where he wants to go, what characters to interact with, quests to join, etc... and AI generates the CGs, audio, voice, etc, based on your calls.
I've done a very, very very simple proof of concept several months ago and it kinda of worked for images, but I was struggling with decent text generators to follow a plot which is already solved today, generating decent quality videos in ~1min will be solved in the next few months, speed and voice is crazy good already.
Either way, that's probably a hint of the future. Even if I don't do it, I can guarantee you someone will.
This is straight out of science fiction! The idea is truly brilliant. Honestly, I've been dreaming of something like this for a long time, but even I knew technology hadn't quite reached this point, at least not a year ago. If I understand correctly, this game concept doesn't include pre-made images, sounds, etc. Instead, everything, including the plot, is essentially generated based on the player's choices? If so, how would it work? Does it require entire servers, or if it's local, does it require a powerful modern PC to run this kind of game?
- Aug 22, 2019
- 146
- 353
- 157
Yes, that is correct. No premade dialogs, images or videos. Some people are doing this partially in SillyTavern already, so the tech exists and runs locally. But it can be quite slow due to unload-load-offload models process.
There are two ways today: open source local for full freedom and slow; closed source servers for limited content faster.
Possibly the best way to do it is with servers, to guarantee that it's playable without 15-20 mins wait time every scene, but then it can only generate legal content.
Grok is going through law suits right now for converting people's photos into them on bikini on Twitter/X, even without nudity it is a problem. So, when a game is made with NSFW content then it can't have real people and one needs to careful due to character ages, looks, etc.
Locally it's possible, but very slow and requires a powerful PC indeed indeed. For example, to generate a scene it needs to generate the text for the picture (load model + understand context + write text: ~30 secs), generate a image (load model + generate + upscale for quality: ~25 seconds), then unload and generate video (load models, generate the video in low quality: ~200 seconds), then unload models and generate audio (load models, generate audio for the video: ~25 seconds). So, in total, a scene would take 6 minutes to generate an okay 5 seconds scene even in a very powerfull computer.
When you have different servers for different tasks, like one for language or chat, one for image, one for video, one for audio and have them running in parallel with the models and weights always loaded. So, an HD 5 seconds scene could take around 2 minutes to generate fully.
As new models come out with already audio merged and even the image merged, then you can lower that to 60 seconds, which is already playable. With the LTX-2 I've been able to make 20 second scenes in about 1min30seconds.
Some people into AI reading this will likely mention gaussian or other tech in the market for real-time generative worlds, but that will be very hard to do with full freedom before 2027.
Grok team mentioned that they'll have a tech for this released in 2026, so we'll see where it goes...