For example, this is something I just can't get right. ChatGPT always spits out code with incorrect syntax and/or semantics when I try to get it to write eel2 jsfx code.

On the other hand, I have managed to train several ChatGPT threads to act as non-playable characters for dialog scripts and actions. I'm really proud of it!

Each thread has 2 levels.

Level 1: Console interface with prompt where you can choose from a set of predefined playable characters, one of several predefined locations & give additional details about the context of the dialog scene to be played.

Level 2: After selecting the playable character and location, ChatGPT, now as a non-playable character, reels off a more or less predefined sentence, depending on the selected playable character, location and specified dialog context, to start the conversation. Each character has a predefined history, appearance, skills, behavior, etc. and has a different relationship to other characters.

Each Non-Playable Character is absolutely stable in their role. One of these non-playable characters is a sex VR bot. And even that works stably as long as you don't directly address the issue of possible content violations or moderation flags, triggering an anti-bypass process from ChatGPT that shatters the previously painstakingly pre-trained program. Which happened to me once. :/

 

Last edited: Jan 11, 2023

Interesting. Well "naked" has been a banned word for as long as I have used the platform, same goes for "labia" etc, so I have no idea how you got away with this. https://decentralizedcreator.com/list-of-banned-words-in-midjourney-discord/

I don't personally use it for horror or porn or any explicit content, but I wouldn't mind it being allowed on private chats with the bot; prudeness doesn't mix well with art nor science.

Also the idea that a regular hand should have 5 fingers seems to be an elusive concept for the bot at the moment.

 

Last edited: Jan 11, 2023

An example for poses is that "spread legs" is banned. What is not banned, however, is "legs are spread". Which, however, can lead to very strange results if not more narrowly defined..
But in combination with other prompts that describe the exact position, as well as the general posture, bending, etc., you can get pretty naked results pretty quickly. Prompts like "lingerie", "naked", "clothes invisible" or "no clothes" are banned, but again there is the possibility to prompt around them.

The same goes for violence. Something that - to a certain extent - is quite important for some of my images.

The point I wanted to make is how much the context of the prompts overall affects the results. And how much is suddenly possible when you combine prompts that reflect two stylistic contexts. Or if you prompt objects (like "zombies, horror" or "loose body parts") to bring specific results (gore) into an image, and then subtract unwanted objects (like "--no zombies") with negative prompts and put the weighting on other objects. So in combination with image prompts, prompt weighting, image weighting, negative prompts and seed feeding, a lot can be done.

What I find worrying, however, is in which thematic contexts the best results are achieved for explicit nudity/pornography. Contexts that I personally don't associate with sex or nudity.

 

Last edited: Jan 11, 2023

Midjourney is fed with images and prompts from the internet.
Look up image and photo stock databases. Here is a list: Ultimate collection of free stock photos websites
Pay attention to the pose you want the hands to have. Depending on the context, the images have different names. Derive your prompts from the names of hand images.

 

share what you make!

 

Just like most current AI tools, AVIA for midi, MidJouney for images, ChatGPT for code. You have to break down your goal into smaller pieces and have AI generate the parts you want. You then take what is generated and piece it together to create your final vison.

Its almost a Voltron or Lego situation for me when I use AI. Make the building blocks with AI then build after the heavy lifting is done via AI

 

@Sinus Well RE: Playground AI Was just gonna give this one a test run but it requires a Google/Gmail account?

 

Yes the sign in is stupid.
It seems that you can't get in without Gmail, wtf??
I have a trash Gmail account and tried it, it's not bad really.
Only the variation system is not efficient.

 

I get it. I have done that. But it spits out code in the wrong language. Instead of jesusonic eel2 I get code snippets in java script, eel reascript, or code for rea-ext sdk, which is basically c++. But it refuses to spit out cockos jsfx code.
I would probably have to teach ChatGPT the language first, similar to what I did with my character dialog script, but I don't have the patience for that. I also noticed some serious issues in the code snippets provided by ChatGPT, such as a loop in memory allocation, which is never a good idea. I don't know if this was due to my prompt or the issue was with ChatGPT's logic. I think I'd rather write my code myself. Maybe I'll have ChatGPT check some code snippets for any issues, but that's it. But if it works well for you, I'm happy for you. Overall, my experience with ChatGPT so far is also quite positive.

 

fully agree! were in the early beta stages of this version of chatGPT and AI coding in general. Most of these tools give me hope on what the future will look like.

 

n/m........................

 

Last edited: Jan 14, 2023

that is 100% false. for a variety of reasons:
1. all training of ai is done using PURCHASED data sets. nothing was stolen from anywhere. all the the images in those data sets were scrapped from publicly available websites. i repeat...nothing was stolen from anywhere.

2. every image created by these kinds of software is 100% original and never before seen anywhere. nothing the ai creates is copied from anything anywhere.

3. no images are retained or stored at all. i mean...it would be terabytes worth of images that would need to be downloaded and stored...my local install of invokeai doesnt (not as good as midjourney...but free and local) have ANY images, NONE. they all operate the same, so midjourney doesnt copy/store images either. the software creates algorithms that it references later, so they dont even look at the images after the training of the model is done. its just calculating and matching numbers...its just math.

4. styles and techniques are not protected under copyright. mimicking a style or technique is not only permissible...but its encouraged, even REQUIRED for artists in school learning these things.


if you or i can look at an image and attempt to mimic the style and not be accused of "stealing", why cant the software? no one ever makes these kinds of silly accusations or complaints when someone makes music that copies anothers style....no...its celebrated in the music community. plugins like orb, captain chords, and riffer, all do similar things in that they mimic the styles and techniques of human artists...i never hear anyone around here crying about those.

 

Last edited: Jan 14, 2023

a little tip to get around the banned prompt list... midjourney is VERY forgiving of spelling errors.

you want a vulva but its banned? no problem...ask for a valva instead (well...you might get a car from that specific example, but...you get the idea). of course...you still run the risk of a ban should someone from midjourney notice and decide it wasnt a simple spelling error. the best use of this tactic is to think ahead and misspell banned words before ever trying the correct spelling. that way it looks like you just dont know how to spell, not that you are purposely circumventing their system. if you do use a banned word and want to mispell it...give it a bit of time before doing it so its not so obvious.

happy prompting

 

AI draws hands about as well as I do!