Sign In
homemodelsimagesvideospostsarticlesbountieschallengeseventsupdatesshop

Ink & Watercolor Hybrid Style from Legacy Doubao for Flux Dev, NoobAI and Illustrious

128

854

1.9k

54

Updated: Jul 4, 2025

stylewatercolorchineseinkmale

Verified:

SafeTensor

Type

LoRA

Stats

108

20

63

Reviews

(20)

Published

Jun 20, 2025

Base Model

Flux.1 D

Training

Steps: 14,040
Epochs: 18

Trigger Words

An image in style of iwhyb

Hash

AutoV2
B3FBB44D14
default creator card background decoration
A1E7's Avatar

A1E7

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.

Primarily male-focused, but may work in other contexts.

Trained on legacy Doubao AI images primarily featuring mature males, landscapes and animals. While prompted by "ink wash style" in Doubao AI, this is more accurately a blend of ink wash, watercolor and multiple styles.

The dataset includes only one female image, so generated females may not consistently show a distinct Chinese facial feature.

v3.0

After multiple attempts, I still cannot achieve the convergence I'm aiming for. I'm still a newbie when it comes to Flux LoRA training, so I'll leave this decent version as v3.0 anyway.

  • I found sometimes An ink wash painting in style of iwhyb seems to gives better results than original trigger word. Feel free to rephrase.

  • Removed some realistic images from the dataset, which should result in a more consistent painting style.

  • To make the style more prominent, I trained it for over 10k steps. This might lead to blurry faces in distant shots.

    • I know this is a bad practice, but I just don't have the mood to train it again atm.

v2.7(TEST VERSION)

  • Attempt to update to Flux ver, may not align with previous version style well.

  • I find it really hard to generate dadbod, or even just remove abs.

v2.0

  • Reverting to Illustrious base since online trainer still not providing NAI base :(

  • Subtle body hair difference.

  • Use with weight around 0.8. Better use with ADetailer / FaceDetailer, or manually inpaint if necessary.

v1.5(Recommended for HomoSimile XL V3.0 users)

  • Updated to NAI base.

  • No trigger word required.

  • Thicker classic ink style, use ~0.6 for v1.0's watercolor vibe.

v1.0(Watercolor vibe)

  • May seems like watercolor. Less stylized than v0.1 but more stable. Weight 1.0 recommended.

  • Increased dataset; small faces may still blur.

v0.1(TEST VERSION)

Trained with images generated from Doubao AI, as I'm trying to mimic its ink wash paint style when generating robust warrior figure.

Trigger word is dbiwpstyle. This is my first LoRA training, and since I'm still pretty new to this and used a small dataset, the results may contain blurry face and bad anatomy, it's recommended to use with weight lower than 0.6 to avoid such.

There might be a future version to address the above issues. I'm just going to leave this beta version here if any adventurer wants to try it out :)