Authentic Dirndl LoRA

Authentic Bavarian Dirndl fashion for KREA2 and ZImage Turbo models.

No cosplay. No carnival costume. Just authentic Bavarian tradition.

🧠 Description

Summer is here ☀️

Volksfest season is in full swing, the beer gardens are overflowing, and Dirndls are everywhere.

So it seemed like the perfect time to release a new version of my Authentic Dirndl LoRA.

After countless hours of testing, refining and generating thousands of images, I'm finally happy to share this new version.

The goal was never to create a "costume LoRA".

Instead, I wanted to recreate the elegant appearance of authentic Bavarian Dirndl fashion while preserving complete freedom over your characters.

Rather than forcing a specific appearance, this LoRA enhances the authenticity of the clothing while preserving the artistic strengths and unique identity of your favorite base model.

The LoRA was trained using over 130 carefully selected real Dirndl references.

Most importantly:

The training dataset contains no faces.

This means the LoRA will not interfere with your character identities or your favorite checkpoint. It simply dresses your character in an authentic Bavarian Dirndl.

The LoRA was trained using the excellent AI Toolkit by Ostris on RunPod (my local PC simply wasn't powerful enough 💻).

🚀 KREA2

KREA2 has quickly become my favorite image model.

Honestly, I haven't had this much fun with an image model since the early PONY days.

The overall image quality is simply fantastic.

KREA2 already has a surprisingly good understanding of traditional Bavarian clothing. However, the default result often looks more like a generic Oktoberfest costume than authentic regional fashion.

This LoRA significantly improves:

authentic Dirndl proportions
realistic fabrics and materials
traditional Bavarian appearance
much more consistent generations
character consistency for storytelling

The last point became especially important during my recent image series.

Being able to generate the same character across dozens of different scenes while maintaining a consistent authentic Dirndl opens completely new possibilities for visual storytelling and character consistency.

✍️ Prompting for KREA2

KREA2 behaves quite differently from older SDXL or Turbo models.

Instead of simply describing clothing, I recommend describing the entire scene.

My prompts usually follow this structure:

Camera
Character
Scene
Wardrobe
World
Light
Style

Feel free to use my published example prompts as inspiration.

One small trick works surprisingly well:

Instead of writing:

White satin apron with the bow tied on the LEFT side.

try separating both elements:

White satin apron.

White satin bow tied on the LEFT side.

Keeping both the apron and the bow in exactly the same color often produces more consistent results.

Another interesting observation:

Permanent character traits work best inside the Character section.

For example:

barefoot
hairstyle
makeup
facial expression

KREA2 seems to treat these as permanent character attributes rather than clothing.

One final observation completely changed the way I write prompts:

The best images usually don't come from describing a pose.

They come from describing a genuine moment.

Instead of writing:

"She is standing..."

try describing what she is actually experiencing.

Natural interactions with the environment, authentic emotions and small everyday moments usually produce much more cinematic and believable images than static poses.

🍺 Authentic Bavarian Detail

The apron bow isn't just decoration.

Traditionally it indicates the relationship status of the woman wearing the Dirndl.

🎀 Bow on the left side → single / open to a relationship

🎀 Bow on the right side → taken / in a committed relationship

Surprisingly, KREA2 understands this most of the time...

...although every now and then the bow still seems to have a mind of its own. 😉

🚀 KREA2 Improvements

KREA2 represents a significant step forward for authentic traditional clothing generation.

One of the biggest improvements compared to previous image models is the interaction between the Dirndl skirt and the apron.

Earlier models often struggled with sitting poses, dynamic movements or unusual camera angles, frequently causing the apron to float unnaturally or intersect with the skirt.

KREA2 now handles these interactions remarkably well, producing much more natural and believable results across a wide variety of poses and compositions.

This allows the LoRA to fully focus on what it was designed for: enhancing authenticity, consistency and traditional Bavarian style.

🔥 NSFW

The LoRA itself was trained exclusively on authentic traditional clothing.

Like most clothing-focused LoRAs, it can also be combined with community-developed KREA2 workflows where appropriate.

⚡ ZImage Turbo Models

This LoRA was originally developed for ZImage Turbo models, especially:

Big Love Zia1
Big Love ZT3

where it still produces extremely reliable and consistent results.

For best results I recommend a prompt structure like:

wearing dirndl,
black bodice,
white blouse with short puffy sleeves,
red apron,
black skirt

Optional details:

choker
ruffled collar
lace trim
silver chains
buttons
zipper
lace details
detailed textures
floral belt buckle

These details help create even more realistic Dirndl variations.

⚙️ Recommended Settings

Weight: 0.8 – 1.0
Optimized for KREA2
Also works perfectly with Big Love Zia1 / ZT3
Compatible with most ZImage Turbo models
No trigger word required

📌 Notes

Authentic Bavarian Dirndl fashion
No costume-style Dirndl
No faces in the training dataset
Character consistency preserved
Excellent for storytelling
Works for both SFW and NSFW generations
Optimized for KREA2 and ZImage Turbo

Search AI Models & Addons

very few results