Gordian Knot
Half Man, Half Animal, All Dumb
- Joined
- Nov 29, 2018
I haven't tried Midjourney, but I've been messing around with Stable Diffusion for a month or so now. I'm no expert, but I'm surprised just how versatile it is. There are many different models available online for almost any style you want: photo-realistic, anime, comic-book, oil paintings etc. You also have ControlNets, which let you do things like define pose of the generated character or fix hands by specifying gestures (e.g. closed fist, OK-sign, 'devil horns'). There are LoRas too, which - if I understand correctly - apply specific concepts to any given model. I've seen LoRas for characters, objects, poses and even art styles. It's wild.Depends what AI you're using to generate the images. With something like Midjourney... yes. Something like 7-8 out of a 10 scale. Stable diffusion? Ooof, that one's I find is more complicated to get good results for. The base would be maybe a 4 or 5. It's less intuitive and relies more on negative prompts. I will say that it heavily depends what style you're looking to generate as well. The aforementioned are disproportionately better at realistic artwork more than something like a an anime-ish style.
Though my favorite functionality of Stable Diffusion is definitely inpainting. You paint with a brush over parts of the image you want changed and the AI will re-generate only those parts according to your description. It's great for modifying images that are just a few details away from exactly what you're looking for.
That being said, I haven't had the chance to use AI generated or modified images in my roleplays yet. I wouldn't raise any issues if my partner wanted to use an image like that, as long as it looked good.