Want to get rid of the annoying adverts? The answer is simple. There are 4 levels of advertising on AnimeOTK:
Level 1: (Your Level, guests and members with 0-9 posts) has maximum advertising. Level 2: (A user with 10 or more posts who has been a member over two weeks) have vastly reduced advertising (including removal of the large full screen closable adverts) Level 3: (A member who has been registered over a month and has over 50 posts, or has a special account like "artist" or "writer" sees minimal advertising. Level 4: (VIP Donators) receive no advertising.
All donators of $10 or more will never see an advert on our site again!
This is how I did the comic. "Rolling gacha" it's how I've been calling generating a lot of pictures with the AI hoping for some good result.
The spanking panel was mostly entirely guided, with me providing a very crude drawing over an AI generated room. If you zoom in, you can see how crude was that drawing. The crudeness was, in part, on purpose: I'm testing how badly you can draw and still have the AI generate professional looking results by drawing over it. And the answer is "very badly".
The refining part works like this: Once I have the crude picture in the img2img module, I draw a mask over the spanker only, and then describe the background and the spanker with tags: indoors, bed, elf, male, sitting, dark skin, white hair, angry etc.
The idea is to let the AI know what the composition IS, and specially what's under the mask. This lets the AI draw over the crude picture, adding details as needed. The key parameter you play with in img2img is called "Denoising strength" on web-ui or just "Strength" on novel-ai's site. This value controls how much liberty the AI takes with your image. My tests show that values at around .5 to .6 are enough to turn bad sketches into professional-looking art with the right tags describing it. For adjustments on an picture that's quite good already, use values at a .2 to .4 range.
Once the spanker was done by the AI, I clean the mask, send the improved image back to img2img and draw a new mask over the spankee, describing the scene and her with the tags. This process gets quite natural after you play with it for a while, and soon you get the hang of what to tag to fix just small details with small masks.
@DrWhoAR Pretty sure Gacha is just an expression of "selecting by chance", and Img2img is the name of a technology that references an input picture and produces a result picture based on the input. Basically you can put something similar to what you want in the system and it will put out something that's different but close to what you want.
And for this post... honestly amazing, you've explained about how to put out a scene where two characters are in a spanking position! I was honestly having much trouble at doing this...
But... it specifically requires mask? So building a local learning system would be necessary, and NovelAI will certainly not do that... So I would guess a two-people spanking would only be good on local designs, and NovelAI won't be able to do that very well, and is instead only good for single person actions.
@DrWhoAR
These are examples of a guided image generation process using Stable Diffusion, an artificial intelligence that generates pictures on demand. You can see for yourself at novelai.net (generating images requires a cheap subscription) or look around for some guides that let you run these models in your own computer. Google "automatic1111 web-ui" and "stable diffusion models list" for more info.
@1063634615
For the time being, you need to run the image generation locally to create two-people spanking scenes (or any kind of more complex scene, actually). Masks are the key for these.
This is indeed a very ambitious and thoughtful work.
But for me, even the "very badly" drawing is still quite difficult for me, it is difficult for me to conceive the characters and plot in the scene, I think I still lack training....or I have aphantasia .
Anyway, thanks for sharing your work and look forward to seeing more of your work in the future!
