ThalisAI/subway-selfie-concept-lora

Subway Selfie — Concept LoRA [Flux]

A concept LoRA for selfie-style photographs of women in urban transit environments like subway and train stations. Captures the mood of underground stations with atmospheric lighting, reflective surfaces, and that distinct subway ambiance. The model follows prompts well and produces images with a mild digital art style.

Gallery

Late night train — pink hair, punk energy	Rush hour platform — businesswoman selfie

Usage

• No trigger word needed — the concept applies automatically when the LoRA is loaded
• Recommended strength: 0.4–0.9, depending on how strong you want the effect
• Flux guidance: 3.0–4.0 (anything from 2.0–8.0 works to some degree)
• Base model: Flux.1 Dev
• Resolution: 1024x1280 or 1024x1024

Tips

• Works best with selfie angles and subway/train station settings
• Describes moody, atmospheric lighting for strongest results
• Lower strength (0.4–0.6) gives a subtler effect; higher strength (0.7–0.9) produces a stronger stylization
• Combine with character descriptions for varied results — hair color, eye color, clothing, and expression all respond well

Example

import torch
from diffusers import FluxPipeline

pipe = FluxPipeline.from_pretrained("black-forest-labs/FLUX.1-dev", torch_dtype=torch.bfloat16)
pipe.load_lora_weights("ThalisAI/subway-selfie-concept-lora")
pipe.to("cuda")

prompt = "A moody selfie-style shot of a woman with long, straight black hair and piercing blue eyes, taking a selfie in a misty, atmospheric subway station with foggy windows and a sense of mystery."
image = pipe(prompt, num_inference_steps=28, guidance_scale=3.5).images[0]
image.save("output.png")

Example Prompts

• A dramatic shot of a young woman with short, spiky black hair and piercing green eyes, taking a photo of herself in a dimly lit, abandoned subway tunnel with crumbling brick walls and flickering fluorescent lights.
• A vibrant image of a woman with dreadlocks and a bright smile, taking a selfie in a lively, bustling subway station with eclectic music playing and a diverse crowd of people rushing to catch their trains.
• A haunting image of a pale-skinned woman with short, choppy white hair and sunken eyes, taking a photo of herself in a deserted, eerie subway station with flickering lights and an air of unease.

Training Details

• Base model: Flux.1 Dev (fp8)
• Network: LoRA rank 8, alpha 16
• Steps: 1,250
• Optimizer: Prodigy (d_coef=2, weight_decay=0.2)
• Precision: bf16
• Training images: 10
• Quality rating: 4.0/5
• Civitai ID: 1536660

Links

• Gallery & Prompts
• Civitai
• All Models

About

Created by Thalis AI — LoRA training and generative AI experiments. Where the lotus-eaters forgot, we remember.

License

This model is released under the FLUX.1 Dev Non-Commercial License.