WAS Dieselpunk

he uses PNGInfo in WebUI he has mentioned but says to comment on images you are curious about and he may help. He also uses custom models often so prompts won't help reproduce results. He often uses a Idun model which isn't on here.

@3DE So in short, the example images are not reliable?

@WAS Since I can't replicate the results, I can not verify if the example images was done with a single generation. Nor can I determine the strength of the embedding itself because you supposedly used many other embeddings to create those example images. I'm sure the style will get across, but it does not actually tell me what it'll look like for the person downloading the embedding. Nor does it say if the example images was tinkered with extensive inpainting, or touched up by photoshop at the end. I just don't know.

In short, the pictures look pretty, but they don't really say how I would be able to achieve something like that. I'm not asking for a checkpoint model, but a prompt for each image would be nice. Demonstrating the embedding with a commonly accessible checkpoint model would also do a lot to showcasing the strength of the embedding itself.

After all, we're not trying to critique your artistry, which is excellent based on the pictures, but we just want a quick idea of how to at least attempt to replicate the results. It does not mean I wont use the embedding though, but I assume that I will get nothing like the example images, and thus will instead rate the quality based on what I first generate instead.

@MiniMage y do u want to replicate?

@WAS Well, that's what I believed 3DE said, and I just took that as the truth because I had no way to verify if it was wrong or not. Upon closer inspection, I don't to see that part of the comment today though, strange. Anyhow, I had no ability verify anything. The directions does not refute this possibility either.

@3DE By verifying the prompts used in any of the examples, I will know that it's indeed as advertised and thus gets a natural 5/5 from me. If the results are not reproducable, you generally feel a little cheated and would thus shift the rating to a 4/5 or a 3/5.

@WAS It's not really about research, let me give you an analogy. Imagine you click on a youtube cooking video because you see the title say "how to cook authentic noodles", the preview image is also of authentic noodles. There's vegetables and meat, it looks really pretty. But when the video begins, you get no information of the ingredients. You get informed of the heat and the type of stove used, but no information about the ingredients to cook the dish besides the noodles themselves. They then cut to few examples of the finished result. The cooking video skips the cooking. The creator of the video can defend himself and state that this is just an example of the cuisine, that this is a matter of taste. That you can add your own ingredients and make your own noodles. What people want is the instructions on what to add and how to prepare them to look like the preview image. That's what made them click it, that's what they wanted to try to make. Without that there is a logical disconnect between what made you click the video and what you get out of it. Annoyance will result, you'll feel a bit cheated.

Generally, it's only when you feel comfortable with it, that experimentation happens.

In your case, you've provided a resource. A good resource, but a resource. The images provided are there to represent your resource, and serves both as examples and advertisement for that resource. The more beautiful they are, the higher rated the resource will be. But if the images can't be replicated, it'll feel like false advertisement. Which will make people rate it with lower values instead. If I judged based on 3DE's results, I would probably rate this resource at 3/5.

@WAS It's not about achieving the exact result, it's about being given the oppurtunity to try. It's when the user fail spectacularly that the user will ask for help. There might not be much for you to do, maybe a rough pointer in the general direction is what you can do after that. After a certain point, you might simply not be able to help. For the user then, they may still be unable to get the exact result, but they'll get slightly closer. Getting the 1:1 isn't that important, if you've presented the instructions and ingredients and they are still unable; that's a skill issue on their end. The lesson isn't "if at first, you don't succeed, don't try", but "if what you're doing right now isn't working, what's wrong? Check the manual and try again."

The thing is that you're marketing a resource to produce a style. You're not selling pencils, you're selling the potential to produce works at a specific quality and style. The quality of which your resource gets rated for. If you were selling art I would expect your works to be presented at an art gallery. I wouldn't ask an artist how they made the painting in that context. After all, that's not the product you're selling then.

I am grateful for your resource, don't get me wrong, but your resource, your product, is a "how to generate the dieselpunk art style, for dummies". If the dummy in the equation is the ai or the user, that's up to interpretation. I'm still capable of guessing, but this how-to for dummies result feels inadequate if the user has to. Thus the rating tends to fall by the users if the promise you marketed to them is not met. That promise is presented with the example image and the thought in the users mind. "That is amazing, I want to do that, I want to make something very much like that thing, that would bring me so much joy". That's just basic marketing and you did it beautifully.

Do I understand art? Not sure. I can tell if something is pretty, and I can tell if something looks off. But I do enjoy creating stuff with people that want me to create stuff, people who want to help others with the joy of creation and given the chance, I would like to help others if I am able.

Regarding the suggestions given thus far, it's not as much of ingredients, as it is the base trigger words for the embedding to do absolutely anything at all. By the analogy of noodles, it's "Apply it to water, add heat, wait", It gets me about third of the prompt of maybe one of the example images. The level of detail of your works implies far more elaborate prompts.

Also, don't worry. I don't expect ai art to ever be invalidated. Legally it's all greenlit. All those licenes artists tend to sign to be able to submit their works digitally, legally, makes the case that anything sufficiently edited will thus be sufficiently transformed and be considered new works. Thus ai by those licenes alone are already in the clear. Granted, there are cases that may be scrubbed regardless, such as for example gettyimages and the like. Worst case scenario is that some of those simply get scrubbed. Removal of those from the system at this point wouldn't put a dent in its development as sufficient material is being submitted by users by the hour. This would easily plug that gap in less than a day. So you don't need to worry about that one.

@MiniMage you seem to know little about laws and copyright. The legality of AI art is very real. That's why 2.x only uses publicly licensed images, and also why they are under a lawsuit for 1.x.

Also it's weird you say it's not about reproducing, and the immediate say it is by "trying".

Like, almost ALL derivatives are based on 1.x, which is the models in question. And then there is the use of other software like CodeFormer and GFPGAN which is research only anc can't be used for commercial uses, because it contains celebrities and real people without consent.

Just distributing a 1.x model is technically unlawful because of the licensed images used to make it, and therefore a derivative of those copyrighted things.

@3DE Well, if I can get like 80% there, I'm already happy. With the current instructions, I get just about 30% there. As Was stated, 100% is impossible. But the closer people get, the more pleased they get. I asked for ways to get closer, Was said no, that's about the summary of our little chat.

@WAS If it's easier to dismiss my criticism as being ungrateful, then go with that if that makes you feel better. Anyhow, I'll drop a score with my review later. We don't really have anything else to add to this little discussion, since you assume that my criticism is out ungratefulness and/or hostility. If I didn't feel grateful for your contribution, I wouldn't have commented at all in the first place, I would've just grabbed what I wanted in silence.