LoRA model for Flux: Iconic Vespa 125 Retro Scooter

The Iconic_Vespa_125_Retro_Scooter_360 is a LoRA (Low-Rank Adaptation) neural model. This model was trained locally using Fluxgym , leveraging the FLUX.1-schnell base model, an author's variant of Flex.1 by ostris (https://huggingface.co/ostris/Flex.1-alpha).

This LoRA model was trained on a fully copyright-protected and legal synthetic dataset consisting of 12 images. The high precision of LoRA generation, achieving up to 98% similarity, significantly outperforms generation without it (50-60%).

Trigger words: You should use "Vintage Vespa" to trigger the image generation.

When generating images, use commands to change angles.

These commands were entered as text triggers into the dataset during the neural network training process:

(Front view. Left view.)

(Front view. Right view.)

(Rear view. Right view.)

(Rear view. Left view.)

(Bottom view. Left view.)

(Bottom view. Right view.)

(Top view. Left view.)

(Top view. Right view.)

(Top view. Left view. Rear view.)

(Top view. Right view. Rear view.)

(Top view. Right view. Front view.)

(Top view. Left view. Front view.)

(In flight)

(The doors are open)

License Information:

License: Apache License Version 2.0

License Link: https://huggingface.co/ostris/Flex.1-alpha/blob/main/LICENSE.txt

Training Parameters from dataset.toml and train.txt:

Resolution: 1440

Batch Size: 1

Keep Tokens: 1

Class Tokens: 'Vintage Vespa'

Number of Repeats: 10

Learning Rate: 8e-4

Maximum Training Epochs: 8

Save Every N Epochs: 4

Discrete Flow Shift: 3.1582

Guidance Scale: 1

Loss Type: l2

Network Dimension: 100

Sample Every N Steps: 50

Images for Data Set: 12

Instructions for Use:

To generate images using this LoRA model, you will need to use the trigger word Vintage Vespa. The model weights are available in Safetensors format, making them compatible with various popular image generation interfaces such as ComfyUI, AUTOMATIC1111, SD.Next, Invoke AI, and Forge