Partially anime style, which tries to cover all kinds of drawings: portraits, landscapes, architecture and much more.
Description
Changes to this version:
More accurate color rendering.
More detailed shadows.
Fixed errors of empty data and incorrect block order (almost).
The eyes have more clearer contours.
Now 2.5D styles are larger than 2D, while using approximately the same color palette and texture. (I recommend using the tags "anime, anime style, anime screencap" for 2D anime style. Also Danburu tags make the image look more like anime).
A greater variety of drawing styles are supported: graffiti, portrait, oil paints, pastel, semi-realistic, 2D anime, 2.5D anime and cartoon styles. Various types of painting and fine art are supported.
The eyes have clearer contours.
Now the 2D anime style has become truly anime, not 2.5D (I recommend using the tags "anime, anime style, anime screencap" for this. Also Danburu tags make the image look more like anime).
A greater variety of drawing styles are supported: graffiti, portrait, oil paints, pastel, semi-realistic, 2D anime, 2.5D anime and cartoon styles.
Now merge consists of:
Now VAE is built into the model (Grapefruit 4.1 VAE).
No embeddings and LoRa were used in the demo illustrations. I advise you to use easynegative, bad-artist, bad_prompt_version2 to get a clearer and anatomically correct picture.
I recommend using DDIM, UniPC and Heun samplers with 25-40 steps.
FAQ
Comments (3)
Any chance of a Lora version for compatibility hopefully based on SD 2.1 768?
I don't do LoRa, but if you want to do it, then most likely you won't have a problem with it
This model dose line art, drawings, vector, icon art really well.
I think this is the best model that dose this, and I tested many models.