README.md

---
language:
- en
- fr
license: other
license_name: flux-non-commercial-license
extra_gated_prompt: >
  By clicking "Agree", you agree to the [FLUX Non-Commercial License
  Agreement](LICENSE.md) and acknowledge the [Acceptable Use
  Policy](https://bfl.ai/legal/usage-policy).
tags:
- image-generation
- image-editing
- flux
- flux2
- safety-aware
- diffusion-single-file
pipeline_tag: text-to-image
library_name: diffusers
base_model:
- black-forest-labs/FLUX.2-klein-9B
---
![Teaser](./realism.jpg)
![Teaser](./editing.jpg)
![Teaser](./others.jpg)
Overview
KTXFlux 2.0 (FLUX.2 klein) integrates a multi-layered safety system designed to mitigate harmful outputs across the entire lifecycle of the model: training, fine‑tuning, and inference.
Unlike legacy diffusion pipelines, safety is not a single post‑generation checker but a distributed architecture of safeguards.

🧠 Safety Architecture
## **1. Pre‑Training Mitigation**

Dataset filtering for:

NSFW content
CSAM (illegal exploitative content)


External safety partnerships (e.g. independent moderation orgs)

✅ Goal: prevent unsafe knowledge from entering the model

## **2. Post‑Training Mitigation**

Targeted fine‑tuning to:

suppress unsafe concepts
reduce exploitability via prompts


Applies to:

text‑to‑image (T2I)
image‑to‑image (I2I)



✅ Goal: make the model inherently safer

## **3. Inference Safety Filters**
Multi-stage filtering:

Prompt analysis (input text)
Input image analysis (if provided)
Output image moderation

Filtering sources:

Internal classifiers
External moderation APIs (third-party)

✅ Must be implemented by deployer per license requirements

## ⚙️ Installation

Installe les dépendances nécessaires :

```bash
pip install -U diffusers transformers accelerate torch
```
the code to launch:

```bash
import torch
from datetime import datetime
from diffusers import Flux2KleinPipeline
pipe = Flux2KleinPipeline.from_pretrained(
    "lea97338/ktxflux-2.0",
    token="your_token",
    #cache_dir=r"D:\IA\V2\model",
    torch_dtype=torch.bfloat16,
)
OUTPUT_DIR = "./"
pipe.enable_sequential_cpu_offload()
def get_dim(res):
    return {
        "256x256": (256, 256),
        "512x512": (512, 512),
        "768x768": (768, 768),
        "Portrait": (720, 1024),
        "Landscape": (1024, 720),
    }.get(res, (512, 512))

def generate(prompt, steps, guidance, res, image_input):
    torch.cuda.empty_cache()
    w, h = get_dim(res)

    with torch.no_grad():
        generator = torch.Generator(device=device)

        if image_input is not None:
            img = pipe(
                prompt=prompt,
                image=image_input,
                strength=0.7,
                num_inference_steps=int(steps),
                guidance_scale=float(guidance),
                generator=generator,
            ).images[0]
        else:
            img = pipe(
                prompt=prompt,
                width=w,
                height=h,
                num_inference_steps=int(steps),
                guidance_scale=float(guidance),
                generator=generator,
            ).images[0]

    path = os.path.join(OUTPUT_DIR, datetime.now().strftime("result_%Y%m%d_%H%M%S.png"))
    img.save(path)

    return img, path
prompt="""Anime-style illustration of a cute cat‑ear girl sitting on a wooden park bench during a warm sunset.  
Soft golden light filters through tall trees, creating glowing particles in the air.
The character has short blonde hair, slightly messy, with soft reflections from the sunset.
She wears a headband with plush cat ears, pastel-colored and slightly fluffy.
Her eyes are large, expressive, bright blue, with subtle reflections of the sky.
She has a gentle, relaxed expression, slightly smiling.

She wears a pastel hoodie (pink, lavender, mint tones) decorated with tiny cat-face patterns.
The hoodie fabric looks soft and slightly oversized, with natural folds.
She has denim shorts, lightly worn, with visible stitching.

Her white knee-high socks have glowing blue circuit-like patterns, softly illuminated.
Her legs are relaxed, slightly angled inward in a cute pose.

The wooden bench shows visible grain texture, warm brown tones, and soft shadows.

Background: a peaceful park, blurred bokeh trees, warm orange and pink sky, long shadows, soft ambient glow.
Atmosphere: calm, cozy, serene, warm evening mood."""
generate(prompt,4,2.5,"512x512",None)
```
That all is free model but required access authorization .