aiconiccompany/jppy-logo-flux-lora

jppy-logo FLUX.2-klein LoRA

Logo-design LoRA for FLUX.2-klein. Trained on 15,815 logos to bias generation toward clean, vector-friendly, geometric brand-mark aesthetics. Production-tested at scale.

File: flux2_klein_logos_lora.safetensors (2.3 MB, low rank) Trained: 2026-04 on RTX 3090, ~7 hours, loss 0.21 → 0.17 Status: LIVE in production at gpu.social

Training data

15,815 logo images, mixed sources (public domain + creative commons), filtered for monochrome / two-color / clean vector quality. Categories: tech, food, finance, e-commerce, healthcare, education, entertainment.

Usage

from diffusers import FluxPipeline
import torch

pipe = FluxPipeline.from_pretrained(
    "black-forest-labs/FLUX.2-klein-base-4b",
    torch_dtype=torch.bfloat16
).to("cuda")
pipe.load_lora_weights("aiconiccompany/jppy-logo-flux-lora")

image = pipe(
    "a minimalist logo for a coffee shop, jppy-logo style",
    num_inference_steps=28,
    guidance_scale=3.5,
    width=1024, height=1024,
).images[0]

Trigger: include jppy-logo in prompt. Recommended weight: 0.8-1.0 Resolution: 1024×1024 for crisp vector-friendly outputs

Tips

• "minimalist" + subject + "jppy-logo style" → clean two-color marks
• Add color hints: "monochrome", "red and white", "navy and gold"
• For typography-included logos: add "with text [BRAND]" — text quality depends on FLUX base, not this LoRA

Limitations

• Best for logomarks, not full brand systems. No multi-element layouts.
• Text rendering inside logos is hit-or-miss (FLUX base limitation).
• Slight bias toward circular and shield silhouettes from training data composition.

About Aiconic

Built by Aiconic — research-grade AI engineering for production. Used in production at gpu.social for automated logo generation.

Contact: hi@aiconic.company · @yukakust

License

Apache 2.0. Free to use commercially. FLUX.2-klein base has its own license — check before commercial use.