It is bad at generating fantasy creatures, especially centaurs. I tried too many prompts and none of them can reliably produce centaurs. There are a few other fantasy species I have not tried, but I guess the results will be the same.

because they all use the same database, if it doesnt exist in https://www.image-net.org/challenges/LSVRC/ then theres no mask for it and your just making a random composite

I still don't know python. Anyone knows what the following files in the scripts folder are? knn2img.py sample_diffusion.py train_searcher.py

It has one fatal flaw: scene and character consistency. If it could conveniently reproduce the same scene from different perspectives without being weird, and it could create the same original character in different scenes without inconsistencies, it would be beating manga artists. But it doesn't have consistency.

>>9813 Lighting is also bad.

>>9694 Have you tried using the "--scale" parameter? By default it is set to 7.5 and it goes from 0 to 50. The higher the value, the closer to your prompt at the expense of image quality.

>>9813 It would take a general AI to beat manga artists, but if a general AI was created, they would have more to worry about than their job.

>>9692 What is there to think about? Where the fuck is that dude's foot? And his hands? The hooves look weird, too. This algorithm definitely shows some kind of "understanding" of what's being asked for, but it doesn't have a "model" (if you will) of what's being asked for. The astronaut has no feet, no hands, and three arms. To general intelligence, it's recognizable, but on close scrutiny, it falls apart. What the fuck kind of Lovecraftian horror am I looking at? I'm probably missing things about the horse, but am too ignorant of horses to comment on them. It's got a way to go. I'm not really into AI programming, but could we at least make an adversarial model so that the result has to match a model of the target? Our three-armed, no-handed, no-footed astronaut wouldn't be generated, because that doesn't match a model of a human? Just a thought, but it's probably stupid shit, cause I don't know shit for AI.

>>9694 I asked for a satyr, and got this. A literal third leg.

>>9694 Model: Stable Diffusion 1.5 Prompt: centaur with two arms and four legs, horse as a centaur, man as a centaur, man joined to a horse, horse joined to a man, man is a horse from the waist down, horse is a man from the neck up Negative prompt: horse head, horsehead, riding, rider, man leg, woman leg, human leg, pants Seed: 939558753 Sampler: Euler Ancestral CFG scale: 14 Sampling steps: 20 Resolution 448x512 It can generate centaurs, you just have to tell it what a centaur is.

>>9692 unstable animation

>You are not allowed to host, finetune, or do inference with the model or its derivatives on websites/apps/etc. If you want to, please email us at contact@dreamlike.art >You are free to host the model card and files (Without any actual inference or finetuning) on both commercial and non-commercial websites/apps/etc. Please state the full model name (Dreamlike Photoreal 2.0) and include the license as well as a link to the model card (https://huggingface.co/dreamlike-art/dreamlike-photoreal-2.0) >You are free to use the outputs (images) of the model for commercial purposes in teams of 10 or less >You can't use the model to deliberately produce nor share illegal or harmful outputs or content >The authors claims no rights on the outputs you generate, you are free to use them and are accountable for their use which must not go against the provisions set in the license >You may re-distribute the weights. If you do, please be aware you have to include the same use restrictions as the ones in the license and share a copy of the modified CreativeML OpenRAIL-M to all your users (please read the license entirely and carefully) Please read the full license here: https://huggingface.co/dreamlike-art/dreamlike-photoreal-2.0/blob/main/LICENSE.md I say we generate kikes and niggers in all kinds of situations. Let's also improve the model at doing so.

i just discovered this site, never thought i would find a SD thread on here, here let me share some of my best quokkas.

>>11340 The video games board is more active, and has had a few threads on this. >>>/v/765010

>>11340 Nice ones. >i just discovered this site How did you get here?