Sign In

[Krea 2/Z-Image Turbo] DHL Tape Dress

16

Updated: Jul 30, 2026

clothingcostume

Download

1 variant available

SafeTensor

66.85 MB

Verified:

Type
LoRA
Stats

242

2

4

Reviews
Published

Feb 10, 2026

Base Model

ZImageTurbo

Training
Steps: 1,500
Epochs: 5
Usage Tips
Strength: 1
Hash
AutoV2
9548B6A300
Trigger Words
DHLTapeCostume, She/He wears a tight-fitting strapless yellow dress printed repeatedly with the red “DHL” logo and the word “fragile,”
default creator card background decoration
Uploads - 37

37

Downloads - 18758

18.8K

Generations - 52031

52K

License:

Apache 2.0
Z-Turbo-ComfyUI-0005.png

DHL Tape Dress

description

  • some short trend in last year or last two years

Trigger Word

  • [Krea 2] She/He wears a tight-fitting strapless yellow dress printed repeatedly with the red "DHL" logo and the word"fragile,"

  • [Z-Image Turbo] DHLTapeCostume, She/He wears a tight-fitting strapless yellow dress printed repeatedly with the red “DHL” logo and the word “fragile,”

Limitations

  • [Krea 2] Sometimes Logo might be replace a entire image logo just prompt more specific, accidentally forgot to tick bucket resolution so it might be get cropped generated image.

  • [Z-Image Turbo] None

Training Details Krea 2 V0.1 Test

Dataset

  • 20 images (synthetic dataset)

parameters

  • resolution = 1024

  • batch size = 1

  • dim,alpha = 32,32

  • optmizer = AdamW

  • UNet LR = 9e-5

  • scheduler = cosine + warmup 5%

  • timestep sampling : krea2_shift 2.0

steps

  • epochs = 5

  • total steps = 1500

tools

  • musubi-tuner (custom with GUI)

  • torch 2.11.0 cu128

  • Runpod RTX 5090 32 GB

Training Details V2

Dataset

  • 20 images (synthetic dataset)

parameters

  • resolution = 1024

  • batch size = 1

  • dim,alpha = 16,16

  • optmizer = AdamW

  • UNet LR = 1e-4

  • scheduler = cosine

  • timestep sampling : flux_shift

steps

  • epochs = 5

  • total steps = 1500

tools

  • musubi-tuner

  • torch 2.8.0 cu128

  • Runpod RTX 5090 32 GB

Training Details V1

Dataset

  • 20 images

parameters

  • resolution = 768

  • batch size = 1

  • dim,alpha = 16,16

  • optmizer = AdamW

  • UNet LR = 1e-4

  • scheduler = cosine

steps

  • epochs = 5

  • total steps = 1500

tools

  • musubi-tuner

  • torch 2.8.0 cu128

  • Runpod RTX 5090 32 GB

*This LoRA is for studying LoRA training with Z Image turbo and Krea 2

*REPOST model is not allow if you found this model please report to me.