1. What is this:
Introducing the Azur Lane Slow Ahead Style Lora! Dive into the captivating world of Azur Lane with this specialized Lora model designed to mimic the beloved art style of the Slow Ahead anime series.
Target Audience:
Perfect for fans of Azur Lane and the Slow Ahead series, this model is tailor-made for character depictions, kitchen items, food, and scenic backgrounds, both indoors and outdoors.
Key Features:
- Seamless integration with Azur Lane characters and scenes, effortlessly bringing your favorite moments to life in stunning detail.
- Exclusive availability only at Civitai.com, offering endless creative possibilities for customization.
- Simple to use—just start your prompt with "Azur Lane Slow Ahead Style" and watch your ideas come to life.
2. How to use:
Don't know how to prompt with AnythingXL or PonyXL. Read this or this or read my version below:
1. AnythingXL: style, special, artist, special (optional), character name, copyrights, quality, meta, rating, tags
Explain:
+ "style": this is the style tag. For this Lora, use "Azur Lane Slow Ahead Style".
+ "special": years (specify a year for that era art style); NSFW (Not Safe For Work)
+ "quality": masterpiece (add this to the prompt will increase the quality)
+ "tags": tags that are used in Danboru, Gelbooru, ...
2. Pony XL v6.0: score_9, source_anime, character name, put anything here, tag_1, tag_2,...
Explain:
+ "score_9": this is a quality tag. You can use the full list for more quality: "score_9, score_8_up, score_7_up, score_6_up" and add these in the negative prompt if the result is not good enough: "score_4, score_5, score_6".
+ "source_anime": this is a style tag. Full list of style tags: 'source_pony', 'source_furry', 'source_cartoon' and 'source_anime' and ratings of 'rating_safe', 'rating_questionable' and 'rating_explicit'.
+ "character name": this is where you write your character name. Pony will recognize a lot of characters but definitely not most or all of them. You can try to add a tag in a line of "Enterprise from Azur Lane" if the AI doesn't know who is Enterprise.
+ "put anything here": I have no clue.
+ "tag_1, tag_2": tags from image platforms like Danboru, Gelbooru,...
3. Extra:
What is this model good at?
- Works well with various subjects, including characters, kitchen items, food, and scenery, with preferred aspect ratios for optimal results.
- Incorporates prompt engineering for AnythingXL / Pony XL v6.0, allowing for specialized prompts and fine-tuning of images.
Notes:
Please be aware that while this model excels at portraying female characters, it may exhibit bias towards such subjects. However, this does not diminish its ability to deliver high-quality, captivating imagery inspired by Azur Lane's universe.
---
The preferred aspect ratio (Width x Height) is
+ 1 : 1 (1024 : 1024).
+ 3 : 2 (1536 : 1024).
+ 3 : 4 (1024 : 1366) or 4 : 3 (1366 : 1024).
+ 16 : 9 (1820 : 1024).Try adding "cat", "duck", or "bird" into the prompt to generate Meowfficer or Manjuu.
Click on the Comparison image in the Pony XL version and view the post for more details.
Findings:
Apparently, you can add something in the line of "Character-name from Azur Lane" to recreate characters that appear in the series with very high similarities.
Example 1:
LoRA
Azur Lane Slow Ahead: weight 1.0
Azur Lane Slow Ahead Style, Taihou From Azur Lane, BREAK white bikini, crouching, lean forward, playing with sand, on a beach, cloudy, 3/4 view, under eyes level, best quality, masterpiece, detailed
Negative prompt: deformed, bad quality, ******* drawing, beginner drawing, deformed, bad anatomy, deformed, white background, deformed hands, deformed feet, ugly eyes, ugly tits, watermark
Steps: 40, CFG scale: 8, width: 832, height: 1216, Seed: 1803905026, Clip skip: 2, baseModel: SDXL, Sampler: Euler a
Example 2:
LoRA
Azur Lane Slow Ahead: weight 1.3
Azur Lane Slow Ahead Style, Taihou From Azur Lane, BREAK white bikini, crouching, lean forward, playing with sand, on a beach, cloudy, 3/4 view, under eyes level, best quality, masterpiece, detailed
Negative prompt: deformed, bad quality, ******* drawing, beginner drawing, deformed, bad anatomy, deformed, white background, deformed hands, deformed feet, ugly eyes, ugly tits, watermark
Steps: 40, CFG scale: 8, width: 832, height: 1216, Seed: 1251735369, Clip skip: 2, baseModel: SDXL, Sampler: Euler a
Read here for the full list of characters from the series using AnythingXL (or PonyXL)
The results are hilarious, some looks like they are from Siren 😂 I expected PonyXL to perform better due to its database. However i think that my AnythingXL version do better even though i used the same dataset. This result may come from my tagging ability for PonyXL :(