Posted on 2023-03-19 22:36:48
Score: 2 (vote Up)

Okabe developed a new device and has decided that Kurisu would be the most ideal test subject.

Posted on 2023-03-19 23:45:52
Score: 1 (vote Up)

Completely defenseless, that's a nice thought~
If that device was real, I'd be a willing test subject...

(With pre-negotiated consent, of course!)

Posted on 2023-03-19 23:46:23
Score: 0 (vote Up)

Surely it won't work! hypnosis isn't real~

Posted on 2023-03-20 11:45:08
Score: 0 (vote Up)

Is there a way to get it to draw her holding a pendulum or pocket watch? Maybe trick it by saying something like, doing a dental exam, holding a flashlight, etc to get the arm raised up?

Posted on 2023-03-20 21:20:25
Score: 1 (vote Up)

AI has come quite far quite quickly so there's a few ways you can do that:
- use a controlnet and sketch the shape of what you want, set the weight to a low value, add the descriptors to your prompt
- paste the item you want overtop a pre-existing generation and inpaint the result
- enable a LORA focused on the concept you want (and train one if it doesn't already exist)

A lot of people think that AI art is just prompt -> image, but there are a bunch of tools you can use to gain more control over what's generated. As an example, this is an openpose controlnet pose I created to guide the AI into creating the second panel (along with others). i.imgur.com/xgRctMv.png

Posted on 2023-03-21 17:33:45
Score: 0 (vote Up)

I had never heard of control nets or lora's, is this the thing at hugging face? Fascinating!

Posted on 2023-03-22 00:05:37
Score: 0 (vote Up)

If you're running the A1111 WebUI, LORA functionality is already included by default. For controlnets, you currently need to install the extension manually and download the individual models. There's a lot to learn about how they work, so if you're interested I'd recommend reading guides or watching tutorials on YouTube.