Hugging Face's logo

AIARTCHAN
/
roughnessPainter_v1.0

Text-to-Image
Diffusers
StableDiffusionPipeline
stable-diffusion
aiartchan
Inference Endpoints
Model card Files Community
roughnessPainter_v1.0 / README.md
Kaeya's picture
Kaeya
Update README.md
64521de
preview code
|
raw
history blame contribute delete
7.13 kB
metadata
license: creativeml-openrail-m
library_name: diffusers
pipeline_tag: text-to-image
tags:
  - stable-diffusion
  - aiartchan

roughnessPainter

civitai

[병합대회] Roughness Painter_창작자를 위한 모델

↑ 한국어 설명

Download

img1 img2 img5 img9

Rougheness Painter_Model for Creators

(Translate from korean)

After Midjourney won the art contest, I used the generative Ai.

Attempted to apply to reducing unnecessary time.

Traditionally, there is a mat painting technique that combines pictures or images to produce images for efficiency

A technique used primarily in the game or movie industry to create demo images when planning scenes.

The focus is on replacing the photographic material that is merged here with the material extracted from Ai, specifically to create the material of a clear subject, such as a character.

Rougheness Painter is used to simplify the process of making my artwork.

All existing models can simplify painting.

For me, it is used as a model for fast processing of character mat painting.

The real-life process is simple.

Rough Sketch -> Create Variation with Ai -> Retouch and Complete

Use the images generated during this process to quickly create thin concept art.

It is created in an easy-to-apply form by creators with a style of painting (e.g., WLOP, NIXEU, or more casual style) called Mute, Painting Style.

Characteristics

  1. Rough and abstract details
  • Recently produced models such as Gunji Mix have very high completeness.

High-quality images are good to use on their own, but it is a loss to retouch in Photoshop and Clip Studio.

A separate painting operation is required to match the imagined image with the implemented image, which is more efficient.

It reduces completeness, but increases the accessibility and efficiency of retouching by capturing the volume and silhouette of the shape.

  1. Image extraction with low steps and scale
  • Low scale adjustment is necessary to obtain the abstract silhouette as much as possible by utilizing the unique rough feeling.

Since the scale has been lowered, steps are also lowered and generation time is minimized.

This is to avoid quality degradation for the purpose of application in this process.

  1. No outline except for subject or face
  • The outline of the subject has the effect of making the shape clear, but I don't think any more outline is necessary.

It should be easy to express depth indirectly by providing a detailed contrast between the subject and the background.

  1. (Most important) Do not rely on additional functions such as rollers and embeddings.
  • Of course, the functions I need can be improved by adding additional functions such as Laura and Embedding.

However, it is important for creators to show their style well, so if they rely on additional functions,

There is a greater chance of losing personality in the description.

To rule this out, certain styles should not be overstated, and even if they are not, they should be abstract.

In addition, my personal wish was not to rely on weights and not to write long prompts.

List of models used in the merge.

The model at the top of the link is used.

https://civitai.com/models/5414/pastel-mix-stylized-anime-model

https://civitai.com/models/5398/dalcefov3animepastelmix

https://civitai.com/models/4468/counterfeit-v25

https://civitai.com/models/5777/ultracolorv4

Merge expression.

https://civitai.com/models/5414/pastel-mix-stylized-anime-model

*(0.5)

https://civitai.com/models/5398/dalcefov3animepastelmix

= = Source A

https://civitai.com/models/4468/counterfeit-v25

*(0.5)

https://civitai.com/models/5777/ultracolorv4

= = Source B

A + B (*0.5)

Prompt

Base Prompt: masterpice, (best quality), ((best detailed)), depth of field,

Nagative Prompt: (worst quality:1.4, low quality:1.4, bad anatomy:1.4, cropped hands:1.2, extra digit:1.2, fewer digit:1.2), lowres, bad hands, text, error, missing fingers, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, artist name, cropped hands, extra digit, fewer digit, (mask), (fat:1.3), (Chubby:1.3), (curvy:1.3), (loli:1.3), animal_ear, huge breasts, ((2girl)), (3girl), (4girl), (squat_down), corset, toes, (3 arms), (3 legs), multiple_views, ((extra_arms)), extra_limbs, (mutated_hands_and_fingers), bad_prompt_version2, (((Multiple_arms))), (4 arms), (4 legs), multiple_limbs, multiple_legs,

Both t2i and i2i are the same.

Set value

VAE: https://civitai.com/models/5414/pastel-mix-stylized-anime-model

Sampler: `DPM++ 2M Karras / DPM++ SDE Karras``

Steps: 10~20

CFG: 4~9

  • Hires. fix

Upscaler: Lanczos / R-ESRGAN 4x+ Anime6B

Denoise: 0.5

Tip (Optional)

  • When changing the sampling step and scale, it is recommended that you change them proportionally.
  • As the angle of view increases, it is recommended to increase the scale figure to prevent shape crushing, but it is not recommended to increase the number above 10 because it is not suitable for the purpose of the model.
  • Embedding can be used for finger detail, but it is not recommended to use additional features such as reference and embedding.
  • Unlike pastel mix, Latent is not recommended. Latent improves the feeling of brush touch, but we need to find the silhouette and the sense of volume, not the brush touch.
  • Lanczos and Anime6B are suitable for retouching and complete image generation, respectively. Anime6B is recommended for use in the middle of artwork production to create a complete version with Lanczos, t2i, etc.
  • If the contrast between the color or shape of the character and the background is small, you can solve it by adding weights to the prompt (depth of field).