Game Character IP | Sky: Children of the Light_V1.0

0
0
0
0
AnimeAnime CharacterGirlCharacter
Recently Updated: First Published:
Anime,Anime Character,Girl,LoRA,SD 1.5Image info
Anime,Anime Character,Girl,LoRA,SD 1.5Image info
Anime,Anime Character,Girl,LoRA,SD 1.5Image info
Anime,Anime Character,Girl,LoRA,SD 1.5Image info
Anime,Anime Character,Girl,LoRA,SD 1.5Image info
Anime,Anime Character,Girl,LoRA,SD 1.5Image info
Anime,Anime Character,Girl,LoRA,SD 1.5Image info
Anime,Anime Character,Girl,LoRA,SD 1.5Image info
Anime,Anime Character,Girl,LoRA,SD 1.5Image info
Anime,Anime Character,Girl,LoRA,SD 1.5Image info
Anime,Anime Character,Girl,LoRA,SD 1.5Image info
Anime,Anime Character,Girl,LoRA,SD 1.5Image info
Anime,Anime Character,Girl,LoRA,SD 1.5Image info
Anime,Anime Character,Girl,LoRA,SD 1.5Image info
Anime,Anime Character,Girl,LoRA,SD 1.5Image info
Anime,Anime Character,Girl,LoRA,SD 1.5Image info
Anime,Anime Character,Girl,LoRA,SD 1.5Image info
Anime,Anime Character,Girl,LoRA,SD 1.5Image info

******************* About Version and Training Set *******************

Currently at version V1.0, this is my own experimental version with room for improvement in generalization. Will update with optimizations when time allows.

Base model used is helloyoung25d_V10f (https://www.liblib.ai/modelinfo/c507b825739e475298f972445b7e2035)

【Trigger Words】 guangyu (optional)

* Training set images are sourced from original game screenshots and partially processed game screenshots.

【Training Set】 589 images  |  Steps: 39540  |  Trainer: Autumn Leaf Integrated Version lora-scripts-v1.8.1

【Disclaimer】

1. Generally, this model is for educational purposes only and should not be used for commercial purposes.

2. Users are responsible for any intellectual property disputes arising from the content generated based on this model, provided they comply with national laws and terms of service..

3. Without the author's authorization, the model service should not be integrated into any third-party software or service.

4. When sharing generated images on social media, credit the model and author in the caption.


******************* Usage Recommendations *******************

【Base Model Usage Recommendations】

Models suitable for anime, gaming, and illustration themes tend to produce better results with flatter facial features, not suitable for realistic models.

Recommended models include:

hello series non-realistic large models (https://www.liblib.ai/modelinfo/045b22166d4248d4a5078d2c74e012c9  /  https://www.liblib.ai/modelinfo/c507b825739e475298f972445b7e2035  /  https://www.liblib.ai/modelinfo/7542fcfbfb0944cb93478ff88f2c8b8c)

Anime/2.5D large models (revAnimated series)

Children's illustration/storybook large models (https://www.liblib.ai/modelinfo/8b4b7eb6aa2c480bbe65ca3d4625632d /  https://www.liblib.ai/modelinfo/093469d326ba4829b26f7df599f81d70  / )

Chibi models (https://www.liblib.ai/modelinfo/33253de83fcb4465a63cd3c421723cf4 /  https://www.liblib.ai/modelinfo/02cbe1df0b5a0e55b4628303c6663fe1 /  https://www.liblib.ai/modelinfo/5bdd3ee7d8f44271909e8c7ae16eca0b)

* The above base models are some recommendations, feel free to experiment based on your preferences~

【Lora Weight Recommendations】

For using anime and illustration themed base models, Lora weight between 0.5-0.8 is ideal; for 2.5D models, a Lora weight between 0.6-0.9 works better. To generate pure environment scenes, include keywords like map/scene/venue/environment/background or increase the weight of specific map names.

* (Due to the specific facial features of certain characters, extreme Lora weights can lead to distortions. To avoid this, local adjustments might be needed if not aiming for the character's specific features).

【Parameter Settings Recommendations】

Image dimensions: 768*768, 1024*1024, 768*1024, 900*1200, etc. (Avoid overly elongated or wide proportions)

Sampling methods: Euler a, DPM++ 2M Karras,

Sampling steps: 24-30

Clip skipping: 2

Redraw range: 0.2-0.5

CFG ratio: 7

VAE: Use as needed


******************* Trigger Word Usage Manual *******************

When using character trigger words, for a closer match to in-game appearances (such as capes and hairstyles), consider using 1 sky kid or sky kids. Using 1boy or 1girl can result in more dynamic character expressions.

The current version includes some hairstyles, capes, pants, maps, and actions, detailed as follows:

1. Maps:

Cloud Wilds: yunye, butterfly plain, floating island

Cloud Peaks: yunfeng, cloud peaks

Rainforest: yulin, rainforest

Twilight Valley: xiagu, Valley of Triumph, Xiaguang City

Cloud Nests: yunchao, Cloud Nest, small town

Golden Wasteland: mutu, Golden Wasteland

2. Capes: various primary capes (use color term + cape), Christmas cape, white velvet cape, colorful scarf cape, pink petals cape, rainbow cape, ranger cape, worn cape

3. Pants: shorts, samurai pants, floor-length trousers, dress, Rhyme Season Pants, blue pants for the Season of Flight, orange pants for the Season of Flight

4. Hairstyles: initial short hair, ponytail, keel hairstyle, wreath, rainforest long hair, Shota, cat ear, kaka, gugu, short hair with bangs, double horn braid, puff ball hairstyle, Ruo Li, green ribbon, Ming hat, witch hairstyle, double twist braids, shoulder-length hair, dreadlock short hair

5. Actions: stand, attract butterflies/catch butterflies, clasping fists salute, hug, hold hands, blow a kiss, dance

* (Increase the weight of trigger words if desired results are not achieved)


Feel free to share feedback and queries, looking forward to your artwork contributions!


Discussion

Most popular
|
Newest
Send
Coming Soon
Download
(0.00KB)
Version Details
Type
Generation Count
0
Downloads
0
Recommended Parameters
Recommended Checkpoint
None
Recommended weight
0.6
CFG
7
VAE
None

Gallery

Most popular
|
Newest