krea2-isometric-3d-lora

Krea 2 DreamBooth LoRA — isometric skeuomorphic 3D

<Gallery />

Model description

An isometric skeuomorphic 3D style LoRA for Krea 2, a Krea 2 re-creation of the popular FLUX LoRA multimodalart/isometric-skeumorphic-3d-bnb. Trained with the Krea 2 diffusers trainer on aisheets/Day_to_Day_Objects_isometric_skeumorphic_3d_bnb.

Training config: rank 32 / alpha 32, attention-only layers (to_q,to_k,to_v,to_out.0,to_gate), learning rate 3e-4 (constant), 1000 steps.

The style is baked in as the default look — no trigger word required — and generalizes well beyond the everyday-object training set to characters and full isometric scenes/dioramas.

Krea 2 ships as two checkpoints: RAW (the non-distilled base you fine-tune on) and Turbo (an 8-step distilled checkpoint for fast inference). Train your LoRA on RAW and run it on Turbo.

Use it with the 🧨 diffusers library

>>> import torch
>>> from diffusers import Krea2Pipeline

>>> # Load the LoRA onto Krea 2 Turbo (the distilled inference model)
>>> pipe = Krea2Pipeline.from_pretrained("krea/Krea-2-Turbo", torch_dtype=torch.bfloat16).to("cuda")
>>> pipe.load_lora_weights("linoyts/krea2-isometric-3d-lora")

>>> # Turbo recipe: 8 steps, no classifier-free guidance
>>> image = pipe("a cozy isometric bedroom", num_inference_steps=8, guidance_scale=0.0).images[0]
>>> image.save("output.png")