conceptart / README.md
Przemysław Mejna
Update README.md
8312174
metadata
license: openrail

Conceptart embedding version 1.0

This model is made for Stable Diffusion 2.0 checkpoint 768

For whom is this model?

This model is targeted toward people who would love to create a more artistic stuff in SD, to get a cool logo, or stickers concepts, or baseline for an amazing poster. For sure as well for concept artists needing inspiration or indie game dev - who might need some assets. This embedding will be useful as well for all fans of bording games/table top rpg-s.

How to use it?

Simply drop the conceptart-x file (where x is a number of training steps) into the folder named embeddings. It should appear in your SD instance main folder. In your prompt just type in: "XYZ something in style of conceptart-x". This is just an example. The most important part is the conceptart-x.

I would recommend you to first try each of them as they all might behave a bit different.

Issues

Currently, the model has some issues. It tends to have grayish/dull colors sometimes. The object's elements are not ideally coherent. The improvements will come with future versions. You might expect them in the following weeks.

The strengths

One of the biggest strengths of this model is pure creativity and out of the box with proper prompting a good quality of output. The strongest part of the model is a good quality improvement with img2img.

I think ofthen the usual workflow will look as following (ideas):

  1. You prompt-craft and create cool designs,
  2. You select ones you like (sometimes smaller objects/elements/designs from the output)
  3. You go to img2img to get more variations, or you select a smaller element that you like and you generate a bigger version of it. Then you improve on the new one up until you are satisfied.
  4. You use another embedding to get a surprisingly amazing output! Or you already have a design you like!
  5. At The same time you might like to keep the design and upscale it to get a great resolution.

Examples

Basketballs with japanese dragons on them:

I have used the one of the outputs, selected the object I liked with the rectangle took in img2img authomatic1111 ui, and went throught two img 2 img iterations to get the output.

Prompt: ((basketball ball covered in colourful tattoo of a dragons and underground punk stickers)), illustration in style of conceptart-200, oil painting style Negative prompt: bad anatomy, unrealistic, abstract, random, amateur photography, blurred, underwater shot, watermark, logo, demon eyes, plastic skin, ((text)) Steps: 30, Sampler: Euler a, CFG scale: 11.5, Seed: 719571754, Size: 832x832, Model hash: 2c02b20a, Denoising strength: 0.91, Mask blur: 4

Anime demons

Just one extra refinement in img2img.

Prompt: colored illustration of dark beast pokemon in style of conceptart-200, [bright colors] Negative prompt: bad anatomy, unrealistic, abstract, cartoon, random, amateur photography, blurred, underwater shot, watermark, logo, demon eyes, plastic skin, ((text)), ((multiple characters)) ((desaturated colors)) Steps: 24, Sampler: DDIM, CFG scale: 11.5, Seed: 1001839889, Size: 704x896, Model hash: 2c02b20a

Cave entrance

Straight out comparison between the different embeedings. At the end result with vanilla SD 2.0 768

Prompt: colored illustration of dark cave entrance in style of conceptart-200, ((bright background)), ((bright colors)) Negative prompt: bad anatomy, unrealistic, abstract, cartoon, random, amateur photography, blurred, underwater shot, watermark, logo, demon eyes, plastic skin, ((text)), ((multiple characters)) ((desaturated colors)) Steps: 24, Sampler: DDIM, CFG scale: 8, Seed: 1479340448, Size: 768x768, Model hash: 2c02b20a

Enjoy! Hope you will find it helpful!