AniGen V3 Guide
After more finetuning of base SDXL models, after taking inspiration from the available open source AI models, we're proud to present the newest iteration of AniGen image generation models - AniGen V3.
Below is a quick guide on what's improved, along with general instructions on how to use the model.
Prompt Format¶
With AniGen V3 we travel back to the prompting techniques of V1, making it more stylistic and refined to create any kind of anime artwork in any anime style.:
Format: <1girl/1boy/1other/...>, <character>, <series>, <artists>, <description tags>, <quality tags>
Anime artwork should be prompted based on the tags available in Danbooru. Refer to some images in that site to get a good idea of tags. Let's see an example prompt and artwork.
Example Prompt: 1girl, kouzuki kallen, code geass, kyabakurabakufu, one-piece swimsuit, blue swimsuit, wet, outdoors, ultra detailed, perfect quality face, anime high quality, amazing quality, very aesthetic,
Let's separate the parts of the above prompt:
- Gender: 1girl
- Character: kouzuki kallen
- Source: code geass
- Artist Style: kyabakurabakufu
- Dress, Setting, and Other Descriptions: one-piece swimsuit, blue swimsuit, wet, outdoors
- Quality Tags: ultra detailed, perfect quality face, anime high quality, amazing quality, very aesthetic
Quality tags such as masterpiece, best quality are automatically added, but adding more quality tags will help increase the quality of the artwork or improve it's aesthetics.
Tip
While following the above order is always best, it isn't a must. You can change the prompt order and still get good results.
Creating original characters is also simple, and it's not a must to include artist tags. Instead you can include description tags or style tags. Here is an example:
Example Prompt: 1girl, solo, looking_at_viewer, black_hair, hair_ornament, gloves, red_eyes, dress, closed_mouth, jewelry, standing, full_body, earrings, black_gloves, indoors, medium_hair, dark_skin, blunt_bangs, fingerless_gloves, black_footwear, necklace, nail_polish, blurry, black_dress, high_heels, bracelet, dark-skinned_female, blurry_background, depth_of_field, armlet, reflection, backlighting, stairs, long_dress, anklet, asymmetrical_clothes, single_bare_shoulder, gold, egyptian, gold_bracelet, masterpiece, best quality, good quality, very awa, very aesthetic, absurdres, newest,
This prompt adds more details, more effects such as adding blurry backgrounds, and specifies a lot of character details.
Quality Tags¶
Quality tags are words which increase the quality of the artwork. The default quality tags are automatically added if the option is kept checked.
Similar to other models AniGen V3 supports common quality tags, including ‘best quality, 4k, masterpiece’, to enhance artwork quality.
You can also add other quality tags of your own such as:
- Extremely detailed eyes and face
- Beautiful detailed girl (for female characters)
- Extremely detailed CG
- Extremely delicate and beautiful
Generating 2 or more characters¶
Using the BREAK tag can separate the prompt tokens into a new block, which we can use for our advantage to generate two or more characters into the same image. Simply write the features of the first character, enter the BREAK keyword and then the second, and so on.
Example Prompt: 2girls, brown hair, blue eyes BREAK red hair, green eyes, dark skin BREAK smile, looking at viewer, outdoors, garden,
You can see how the first girl has the brown hair, blue eyes, while the second girl has the red hair, green eyes and dark skin that we specified in the prompt. Then the common prompt - smile, looking at viewer, outdoors, garden, - is applied to entire image.
Prompt Weights¶
Prompt Weights genrally mean 'how much improtance should be given to that word(s)'. Generally each word you write in the prompt has a weight of 1. But this can be increased by 0.1 by inserting a curly bracket, or decreased by 0.1 by entering a square bracket.
Here is a prompt and a resulting image with and without prompt weights.
Example Prompt: 1girl, black hair, green eyes, shiny skin, glowing skin, golden necklace, upper body, cowboy shot, looking at viewer, adjusting hair,
Example Prompt: 1girl, black hair, green eyes, {{{shiny skin, glowing skin}}}, [[[[golden necklace]]]], upper body, cowboy shot, looking at viewer, adjusting hair,
As you can see, the second image gives more emphasis to the shiny skin and glowing skin while giving less emphasis to the golden necklace. This is because we increase the weight of shiny skin, glowing skin by 0.3 (with three brackets), and decrease the weight of golden necklace with 0.4 (with four brackets).
Styles¶
Unlike V2, AniGen V3 is very good with adapting to styles. While we have limited realistic generations to prevent abuse, we have no limits on the styles you could apply for anime generation. To use the provided built in styles, see the new UI guide of ours. This is about creating your own styles.
For styles, use natural language as much as possible. If you want to take the full effect of the styles you're applying we recommend not using quality tags or default negatives and instead applying your own quality tags and negatives.
Example Prompt: 1girl, mature, crown, sleeveless dress, looking at viewer, adjusting hair, wind, hair in wind, closeup, outdoors, renaissance painting style, water color, extremely detailed eyes and face,
AniGen V3 works very well with the default styles provided by us as well. For example, if you want to create a manga artwork.
Example Prompt: 1girl, solo, long hair, black hair, alleyway, looking down, horror, shadows, night, + Apply manga style from styles tab on top left
We hope you enjoy the new model and create great ai artwork with it! We will update this guide as more features are introduced. Happy prompting!