GAG - RPG Potions | LoRa XL版本v0.1 LoRA XL CivitAI ver (ID: 197256)

GAG = Game Asset Generator
? Date: 2023-10-23T12:19:42

FF_Potion_Generator ?⚗️✨

• Resolution: 1024x1024

• Architecture: stable-diffusion-xl-v1-base/lora

• Network Dim/Rank: 32.0

• Alpha: 1.0

• Module: networks.lora

• Learning Rates:

  • Overall: 0.0005

  • UNet: 0.0005

  • Text Encoder: 5e-05

• Optimizer: bitsandbytes.optim.adamw.AdamW8bit(weight_decay=0.1)

• Scheduler: cosine_with_restarts

• Warmup steps: 0

• Training Details:

  • Epoch: 15

  • Batches per epoch: 250

  • Gradient accumulation steps: 1

• Images:

  • Train images: 1000

  • Regularization images: 0

• Noise & Regularization:

  • Multires noise iterations: 6.0

  • Multires noise discount: 0.3

  • Min SNR gamma: 5.0

  • Zero terminal SNR: True

  • Max grad norm: 1.0

  • Clip skip: 1

• Dataset: 1 Directory with 500 images

• UNet Weights:

  • Average magnitude: 2.9637

  • Average strength: 0.0109

• Text Encoder Weights:

  • (1) Magnitude: 1.7253 | Strength: 0.0086

  • (2) Magnitude: 1.7251 | Strength: 0.0067

? Tags:

potion, painting of one health potion, potion of healing, health potion, magic potions, magical potions, alchemy concept, hyper realistic poison bottle, rendered art, making a potion, potions, fantasy game spell symbol, rpg game item, alchemist bottles, rpg item render, stylized game art, game icon stylized, hyperdetailed scp artifact jar, 3 d icon for mobile game, 3 d render stylized, casting a spell on a potion ... and 904 more!

Starting my GAG Serries ? ?

This one is a Potion LORA for 2.1 SD models.

Depending on the model you will get quite different results.

You can use a background remover while generating and get:

again this is a test LORA and is heavily dependent on the input model.

LyCORIS version will be up for testing in a few hours.
And a 2.1 Model is expected in 48-60 hours

For PIXALISATION try:
PIXHELL by R6SPY ?

PS: so far you can use it for demonstration purposes (preview version)

with version 1.0 above (lora and models of GAG series)
two versions will be available:
Licensed - can be licensed for in-game usage upon request (using only my work and licensed images from my end)
Community - using sourced free images and sources but you are responsible is used commercially in your project ¯\_(ツ)_/¯ ? ?

描述:

Conducting an initial test with the Civitai LoRA trainer on a limited batch.

Date: 2023-10-23T12:19:42 Title: FF_Potion_Generator

Resolution: 1024x1024 Architecture: stable-diffusion-xl-v1-base/lora

Network Dim/Rank: 32.0 Alpha: 1.0

Module: networks.lora

Learning Rate (LR): 0.0005 UNet LR: 0.0005 TE LR: 5e-05

Optimizer: bitsandbytes.optim.adamw.AdamW8bit(weight_decay=0.1)

Scheduler: cosine_with_restarts Warmup steps: 0

Epoch: 15 Batches per epoch: 250 Gradient accumulation steps: 1

Train images: 1000 Regularization images: 0

Multires noise iterations: 6.0 Multires noise discount: 0.3

Min SNR gamma: 5.0 Zero terminal SNR: True Max grad norm: 1.0 Clip skip: 1

Dataset dirs: 1

500 images

UNet weight average magnitude: 2.9637495446847635

UNet weight average strength: 0.010910434567277522

Text Encoder (1) weight average magnitude: 1.7253436262339745

Text Encoder (1) weight average strength: 0.0086376313085263

Text Encoder (2) weight average magnitude: 1.7251399575500794

Text Encoder (2) weight average strength: 0.006686976993287057

训练词语: