README.md

---
license: cc-by-nc-nd-4.0
base_model:
- black-forest-labs/FLUX.2-klein-base-9B
library_name: diffusers
---

## Overview

This LoRA is published **exclusively for debugging and evaluation purposes**,
to investigate a potential LoRA strength scaling issue in ComfyUI.

## Test Prompt

```text
evospazio, SUV, front three-quarter view, full vehicle in frame,
correct proportions, standard wheelbase,
clean studio lighting, neutral background
```

### 3️⃣ Reference image (training dataset)
![_MG_7173](./images/_MG_7173.png)

## Training Reference Image

The following image is taken directly from the training dataset and shows
the intended vehicle geometry and proportions learned by the LoRA.

## Inference Comparison

The images below were generated using the **same prompt, seed, sampler, steps,
CFG, resolution, and LoRA strength**.

## Strenght 1.0

Base model generation
![Base model generation](./images/base_generation-1.png)

Distilled model generation
![Distilled model generation](./images/distilled_generation-1.png)

## Strenght 3.0

Base model generation
![Base model generation](./images/base_generation-3.png)

Distilled model generation
![Distilled model generation](./images/distilled_generation-3.png)

At LoRA strength = 1.0 the trained vehicle geometry is barely visible.
Only at strength ≈ 3.0 does the result match the training reference.

## Usage Notice

This LoRA is published **solely for debugging and evaluation purposes**
in the context of ComfyUI and related tooling.

Commercial use, redistribution, fine-tuning, merging, or derivative works
are **not permitted**.

If you need to test this model for debugging purposes, you may do so,
but the generated outputs must not be used commercially.