My first LoRa for Pony.

Blonde futa.

[WLD-Core (v.5)]

Designed as a more natural and flexible counterpart to BLD-ESS.

This LoRA is trained on the same dataset as ESS, but uses a less constrained tagging approach, preserving more of the original signals and subtle variations within the data.

Instead of heavily guiding the output, WLD-Core allows the model to behave more freely, resulting in richer expression, more organic detail, and a slightly less “polished” but more natural look.

Despite its name, the “Wild” aspect does not mean instability. Rather, it reflects a reduced level of constraint — enabling the model to adapt smoothly across different prompts, LoRA weights, and checkpoints.

In practice, WLD-Core remains highly stable while offering greater flexibility and responsiveness compared to ESS.

Recommended Settings

General Use

• LoRA strength: 0.3–0.6

• CFG: 4–6

Balanced results with natural variation and expressive detail.

Illustration Checkpoints

• LoRA strength: 0.3-0.6

• CFG: 4–7

Stronger stylistic response with retained structural stability.

For Explorers

WLD-Core responds more noticeably to changes in weight and CFG compared to ESS.

Small adjustments can lead to visible differences in texture, lighting response, and expressive detail.

If ESS represents controlled refinement, WLD-Core represents controlled freedom.

Both are designed to complement each other — choose depending on whether you want precision or flexibility.

[ESS(v.4)]

This LoRA was primarily designed to bring out the full potential of realistic checkpoints, especially Dreamisoa reALmix.

During development, I initially considered updating the latest BLD, but testing revealed that this LoRA also performs exceptionally well when paired with illustration-style checkpoints.
Because of this versatility, I decided to release it as BLD-ESS instead.

ESS stands for Essence or Essential.
The name reflects my hope that this LoRA will become a small but meaningful component in your creative workflow—something flexible, reliable, and easy to build upon.

Recommended Settings

Realistic Checkpoints

• LoRA strength: 0.5

• CFG: 4–6 (up to 7 depending on the prompt)

Stable, well-balanced results with natural textures and eye rendering.

Illustration Checkpoints

• LoRA strength: 0.7

• CFG: 4–7 (more flexible depending on style and intent)

Allows stronger stylistic influence while maintaining structural stability.

For Explorers

These values are starting points, not limits.
Feel free to push strength or CFG slightly higher or lower to explore different balances between realism, stylization, and texture response.
Some prompts benefit from breaking the “safe zone.”

[v.3]

I've tweaked it to better suit my preferences. I'm generating several samples with my check point “Dreamisoa Remix. I'd be happy if you could choose and use them according to your preferences.

[v.2]

Change to avoid raising the chin too much.

Hopefully the shape of the nose has improved.