TESTING MODELS: Papercut

Artificial intelligence has made significant advancements in recent years, and one area where it has had a significant impact is in the field of digital art. Even though Stable Diffusion is mind-boggling in and of itself, this model called Papercut posted by fictiverse on Civitai is even more so. 

I spent the last few days testing this model and playing around with it. I wasn't sure what to expect from it at the beginning. There are only a few pictures on the download page, and they looked great, but my main thought was that papercut would be very limiting in terms of subjects. 

I couldn't be more wrong. This model is just amazing and super creative in terms of creating the actual cuts. But it's not very easy to use, as it requires some "work" to get amazing papercut images.

Don't get me wrong, you can get some great results just by using the keyword "papercut," but the thing is that this model is capable of much more than that. 

With some research into terms and keywords to use in the prompt, as well as fine-tuning the sampling method, this model enters beast mode, paper-cutting Legendary Asian level 10K.

Let's see, first i wasn't sure what i was going for so I started with a simple prompt:

Lion King of the Jungle 

I wasn't testing nothing specific, so I just played around with different configs and samplers, and like I said before, it gives you real nice output right 'out of the box'.

What about a duck? Not just a random duck...
lets get a bit more specific with this one now. So I ask for a:

Philippine duck, 3D graphics with simplistic polygon shape

And again, I got real great output just playing around with some configurations. 

This was the point at which I became completely addicted to this model; I opened Pandora's box... ok, right! It was just Google, but it's almost the same, right? So I entered the papercut rabbit hole in search of keywords I could use to refine my prompt...

From techniques to various artists, DeepCut, an AI model that has gained attention for its ability to produce images that resemble papercut art. I am not sure if it in any way relates to this SD model or not, but I learned that DeepCut uses a technique called "kirigami," which is a variation of papercutting that involves folding and cutting the paper to create a three-dimensional effect. This allows the model to create images that have a sense of movement and depth, which is just not possible with traditional papercut art. The images produced by DC are highly detailed and possess a level of intricacy that is not possible to be achieved by human artists.

One of the other unique features of the Deep Cut model is its ability to generate images that have a sense of depth, similar to that of a shadowbox frame. This is achieved by using a technique called "silhouetting", where the model creates a cutout image by removing parts of the original image, leaving only the subject in the foreground in full color.  

Overall, the Deep Cut model is an exciting development in the field of AI and digital art. Its ability to produce images that resemble traditional papercut art is truly impressive.

Yulia Brodskaya, Hans Christiansen and Joanna Wirażka are all amazing papercut artist, and I am definitely trying them out in my prompts also.  

Lets look at our keyword list now:

3D graphics with simplistic polygon shapes, intricate designs, delicate patterns, kirigami, shadowbox frame, depth, silhouetting, cutout, exciting, traditional, impressive, ((art by Yulia Brodskaya, Hans Christiansen and Joanna Wirażka)), ((papercut))

Those more specific term related to papercut can, of course, be combined with others that are typically effective with SD prompt engineering. There is a great list at the git repository about the dynamic-prompts extension. 

But be aware, there can definitely be "too much of a good thing." 

It's like cooking. If a cook who is going to make a meal just grabs everything out of his full fridge and puts it in a pot and cooks the hell out of it, it is very likely that the result would be a dish that is unappetizing and not very yummy. 

The combination of ingredients would likely not complement each other, and the flavors and textures would be totally unbalanced. So, the cook needs to carefully select ingredients when cooking to ensure a delicious and enjoyable meal. 

It's basically the same with prompting: too many details and the AI gets "confused," too few details and the result isn't what you expected... finetuning is everything.

Here are some results with prompts using combinations of those keywords and trying out different sampling methods.

((Papercut)), underwater scene, Common Lionfish, 3D graphics with simplistic polygon shapes, intricate designs, delicate patterns

((Papercut)) small boy going fishing on a boat, shapes, intricate designs, delicate patterns, shadowbox frame, silhouette, kirigami

And some others just for show next. The prompts are gone, but using the above list of keywords you should get results that are also amazing.

Some may say that I have spent way too much time with this, but I say that I still have much to test about this model. 

I was totally surprised with this papercut model and still can't believe how wrong I was about it in the beginning. It turned out to be very capable of making papercuts from very detailed subjects and delivering very pleasant images. 

A big thanks from me to fictiverse for uploading this amazing model.

You can download Papercut and other great models from Civitai.

cat tax