Sign In

Style of Kotaro Chiba 🌊 Flux.1-D

12
55
0
6
Updated: Feb 19, 2025
styleukiyo-ekotaro chiba
Verified:
SafeTensor
Type
LoRA
Stats
55
0
Reviews
Published
Oct 11, 2024
Base Model
Flux.1 D
Training
Steps: 3,250
Usage Tips
Strength: 1
Trigger Words
In style of Kotaro Chiba
Hash
AutoV2
A738921549
seruva19's Avatar
seruva19
The FLUX.1 [dev] Model is licensed by Black Forest Labs. Inc. under the FLUX.1 [dev] Non-Commercial License. Copyright Black Forest Labs. Inc.
IN NO EVENT SHALL BLACK FOREST LABS, INC. BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH USE OF THIS MODEL.

Overview

This LoRA is an effort to capture the style of Kotaro Chiba, a modern Japanese artist known for his stunning illustrations inspired by the Ukiyo-e art style, which thrived in Japan from the 17th to 19th centuries. His work is praised for its intricate detail, traditional motifs, and elaborate scenes featuring female characters, animals, and mythical creatures in traditional Japanese attire. Seamlessly blending traditional elements with modern techniques, Chiba brings a fresh, contemporary perspective to Ukiyo-e, offering a captivating fusion of past and present while preserving its cultural essence with a distinctive oriental flair.

Usage

Showcase images were generated in ComfyUI with the following parameters:

Model: flux1-dev (fp8e4m3fn)
Text Encoder: t5pxxl_fp16
Sampler: euler
Scheduler: 24 steps (normal)
Flux Guidance: 4
LoRA Strength: 1

Training

LoRA fine-tuning was conducted using 274 images, which were auto-captioned with InternLM-XComposer2-4KHD. The prompt used was "<ImageHere>Please describe this image in detail." All captions were prefixed with the phrase "In style of Kotaro Chiba." AI Toolkit was utilized (on a single RTX 3090) with the following hyperparameters (for parameters not mentioned below the default values were left):

Rank: 16
Alpha: 1
Optimizer: prodigy
Steps: 3250
Batch size: 1
Learning rate: 1
Learning rate scheduler: cosine
Decouple: true
Use bias correction: false
Betas: (0.9, 0.99)
Weight decay: 0.05
Noise offset: 0.1
Resolution: (512, 768, 1024)