Quercus robur — Pedunculate Oak LoRA for Krea 2

A botanical LoRA for generating Quercus robur (pedunculate oak / English oak) with Krea 2.

The LoRA was trained to improve the characteristic appearance of the species across a variety of views, ages and seasons. The training dataset contains 99 carefully curated and individually captioned reference images covering whole trees, solitary trees, woodland settings, crowns, trunks and bark, branches, leaves, acorns and catkins.

Trigger

qrobur

The trigger can be combined freely with ordinary descriptions, for example:

qrobur, an ancient solitary oak tree in a meadow

qrobur, a leafless oak tree in winter

qrobur, a close-up of deeply furrowed oak bark

qrobur, oak leaves and acorns

Intended use

The LoRA is intended for realistic depictions of Quercus robur in landscapes, nature scenes and botanical imagery. It can be used for complete trees as well as closer views of characteristic features.

It was trained as a species concept rather than a particular individual tree, location or photographic style, so prompts remain free to control weather, season, surroundings, lighting and composition.

Strength: 1.0

Training

Base model: Krea 2

Dataset: 99 curated images

Resolution: 1 MP

LoRA rank: 32

Alpha: 32

Learning rate: 1e-4

Optimizer: AdamW 8-bit

Training steps: approximately 2,475

The published checkpoint was selected from epoch 25. Later epochs produced stronger species characteristics but also began to influence unrelated surrounding vegetation more noticeably.
All showcase images are direct Krea 2 outputs using the LoRA, without external refinement or post-processing.

Trained on an AMD Radeon AI PRO R9700 32 GB using Fizgig ROCm on Windows.