flux-islamic-parametric-lora

FLUX.1-dev LoRA — Islamic Parametric Architecture

Adapter for Islamic Parametric Architecture & Facades: geometric lattice screens, mashrabiya parametric patterns, and structurally symmetric modern envelopes. Geometry is evaluated with measurable metrics (symmetry, edges, periodicity) — not a hallucination guarantee.

License: non-commercial research/demo only (FLUX.1-dev + derivatives non-commercial).

Trigger words

• Primary trigger (required): in Islamic_Parametric style
• Style tag: Islamic_Parametric

Example prompt:

A modern cultural center facade in Islamic_Parametric style, geometric parametric wooden panels, realistic lighting, 8k architectural photo

Training metadata

Usage

import torch
from diffusers import FluxPipeline

pipe = FluxPipeline.from_pretrained("black-forest-labs/FLUX.1-dev", torch_dtype=torch.bfloat16)
pipe.load_lora_weights("Shehab-Hegab/flux-islamic-parametric-lora", weight_name="pytorch_lora_weights.safetensors")
pipe.enable_model_cpu_offload()
image = pipe(
    "A contemporary mosque exterior in Islamic_Parametric style, mashrabiya lattice screen, golden hour, photorealistic 8k",
    height=1024, width=1024, num_inference_steps=28, guidance_scale=3.5,
).images[0]
image.save("out.png")

Architectural design intent

The adapter is trained on a curated set of modern Islamic parametric facades so generated views preserve modular rhythm, bilateral symmetry, and lattice topology — evaluation grids in the training repo compare base FLUX.1-dev against this adapter side-by-side at fixed seed 42 to verify geometric structural logic.

Evaluation

See inference_eval.py (base vs LoRA grids, seed 42) and evaluate_structure.py (bilateral symmetry error, edge density, Hough-line consistency, periodicity) in the source repository. Prompts are locked to the trigger word.