ArtificialGuyBR

Home / Blog / Fixing things

Snapchat Selfie LoRA for FLUX.2 Klein 9B

Generate authentic Snapchat-style selfies with FLUX.2 Klein using this LoRA. Natural lighting, casual framing, Apache 2.0 licensed.

By ArtificialGuyBR Published Updated 2 sources cited Fixing things

TL;DR

  • LoRA that adds Snapchat selfie aesthetic to FLUX.2 Klein 9B
  • Trigger words: Snapchat Selfie or Snapchat pic.
  • Apache 2.0 license, safetensors format, ~993 Civitai downloads

What This Model Does

This LoRA teaches FLUX.2 Klein 9B to generate images that look like genuine Snapchat selfies — candid front-camera shots with natural window lighting, casual indoor backgrounds (bedrooms, living spaces), authentic expressions, and the slight wide-angle perspective of a phone held at arm's length. It's not a beauty filter or a glamour preset; it's the specific "just took a selfie" look.

Example Snapchat selfie style — young woman with wavy brown hair, white t-shirt, warm window light from left, bedroom background

Training & Base Model

  • Base model: black-forest-labs/FLUX.2-klein-9B (the 9B parameter distilled variant)
  • Training compute: GPU time sponsored by Redmond.AI
  • Training data: Curated Snapchat-style selfie images
  • Format: Safetensors
  • License: Apache 2.0

Trigger Words

Include either phrase in your prompt:

  • Snapchat Selfie
  • Snapchat pic.

The instance prompt used during training was Snapchat Selfie, Snapchat pic. — using both together works well.

How to Use

  1. Download the .safetensors file from Hugging Face or Civitai
  2. Place in ComfyUI/models/loras/
  3. Add a Load LoRA node, select the model, set strength 0.8–1.0
  4. Include trigger words in your positive prompt

Diffusers (Python)

from diffusers import FluxPipeline
import torch

pipe = FluxPipeline.from_pretrained(
    "black-forest-labs/FLUX.2-klein-9B",
    torch_dtype=torch.bfloat16
).to("cuda")

pipe.load_lora_weights("artificialguybr/SNAPCHATSELFIE-REDMOND-FLUXKLEIN")

image = pipe(
    "Snapchat Selfie, young woman, wavy brown hair, white t-shirt, bedroom, warm window light",
    guidance_scale=3.5,
    num_inference_steps=28
).images[0]

Parameter | Value

|---|---

LoRA strength | 0.8 – 1.0 Guidance scale (FLUX) | 3.0 – 4.0 Steps | 20 – 30 Resolution | 1024×1024 or 512×512 Sampler | euler / dpmpp_2m Second example — similar casual selfie aesthetic with natural lighting

Example Prompts

Basic selfie

Snapchat Selfie, young woman, long wavy brown hair, white cotton t-shirt, bedroom background, warm afternoon window light from left, natural skin texture, candid expression

With environment

Snapchat pic., mirror selfie, bathroom lighting, casual hoodie, wet hair, phone visible in reflection, authentic morning vibe

Group selfie

Snapchat Selfie, two friends, arms around each other, laughing, golden hour outdoor, slightly tilted angle, genuine moment

What the Images Actually Show

Inspecting the sample outputs confirms the model delivers on its promise: the three reference images all depict young women in casual clothing (white t-shirts), photographed in bedroom settings with warm directional window light from the side. Skin texture is natural — pores and subtle imperfections visible, not airbrushed. Expressions are relaxed, not posed. The framing matches a front-facing phone camera held at chest height. This consistency across samples suggests the LoRA successfully captured the target aesthetic.

Third example showing consistent style — casual selfie with natural window lighting

Common Errors and Fixes

Error | Cause | Fix

|---|---|---

Style not activating | Missing trigger words | Add Snapchat Selfie or Snapchat pic. to prompt Over-cooked look | LoRA strength too high | Reduce to 0.6–0.7 Wrong base model | Using FLUX.1 or other variant | Must use FLUX.2-klein-9B exactly Blurry/low detail | Too few steps | Use 25+ steps for FLUX

FAQ

Q: Does this work with FLUX.1-dev or FLUX.1-schnell? A: No. This LoRA was trained specifically on FLUX.2 Klein 9B. Using it on other base models will produce unpredictable results.

Q: Can I use this commercially? A: Yes. Apache 2.0 permits commercial use with attribution. Include the license notice in distributions.

Q: What LoRA strength should I start with? A: 0.8 is a safe default. Increase to 1.0 if the style is too subtle; decrease to 0.6 if it overwhelms the prompt.

Q: Where can I find more models by this author? A: artificialguy.com — portfolio, prompt library at FindGoodPrompt, and AI tools at JustGoodAI.

Sources

  • Hugging Face model card: https://huggingface.co/artificialguybr/SNAPCHATSELFIE-REDMOND-FLUXKLEIN
  • Civitai model page: https://civitai.com/models/2424835
  • Redmond.AI (compute sponsor): https://redmond.ai/
  • Author's portfolio: https://artificialguy.com/