I'm pleased to introduce my new LoRA model!

This is an experimental model that has already proven to be highly flexible, delivering varied results based on the parameters and techniques used.

The model was trained on 96 images at a 768x768 resolution, all entirely generated from multiple style transfers without the use of any other LoRA model. The captions are descriptive, generated using VLM, so the model responds exceptionally well to highly descriptive prompts.

The trigger word is soft_x, but I recommend adding other keywords such as 3D, smooth, 3D icon, 3D render, and plastic to enhance the effects.

The achievable effects vary mainly depending on input and output resolution, the number of steps, and the sampler. The most interesting effects are achieved using the LCM sampler with 15 steps.

With LCM, a LoRA weight of 1 is sufficient; for other samplers, you can experiment with slightly higher values, though I wouldn’t recommend exceeding 1.5.

The blurred effect you may notice isn’t a flaw; the model was trained on images with a subtle blur to reduce contrast and create a "foggy" effect, generating softer tones.

The sample images you see were generated through various workflows, both Image2Image and Text2Image.

You can use Image2Image to transfer the style onto any image while retaining the original colors. The model performs exceptionally well in this mode.

There’s a lot to experiment with and explore in this model, and I’m eager to see how you’ll use it and what you’ll create.

Enjoy!