Krea 2 - Recursive Fractalism

This LoRA introduces the "Recursive Fractalism" style specifically for the local Krea 2, enabling the creation of strong mind-bending, self-similar, and recursive spaces from simple prompts without vanilla very hard prompting, regional adjustments, inpainting or ID4's JSON magic and high requirements.

The elements can repeat and loop into themselves, forming infinite, spiraling structures of architecture and figures / bodies.

The training involved a carefully selected set of images normalized to 2K-resolution, ensuring the model captures the fine, minute details essential for complex fractal patterns, even at the cost of ~10x longer training time.

For best results, pair this LoRA with painterly, anime or illustrative prompts. While it's compatible with a variety of styles, the recursive effects are most consistent and visually compelling when not pushed toward strict photorealism. Also, see the lora compatibility Warning at the end of this text.

Technical Instructions:

1. Apply the Recursive Fractalism (this) LoRA to the Krea 2 base (Raw) model or use hybrid raw-turbo workflow.

2. Use the trigger word "style: Recursive Fractalism." in your prompt.

3. If you want to replicate my workflow, also include the auxiliary Krea 2 turbo LoRA in your setup. If using ComfyUI, ensure the KJNodes pack is installed, they introduce the model loader and the prompt weighter (not used), and you can use native components instead.

4. For a complete setup, refer to the provided workflow in the images or from my Github link.

If you are reading this from the dotcom site and you have less than 7 images, you can check out the red counterpart, because some examples are anatomically "messy" (though SFW).

The full prompt format is:

style: Recursive Fractalism.
subject: ...

E.g.

style: Recursive Fractalism.
subject: An anime girl.

Or

style: Recursive Fractalism.
subject: A swarm of cats.

Tips for Getting the Best Results:

1. The style trigger word "style: Recursive Fractalism." is self-contained and you don't need to add any more descriptors to this section.

2. The recursive effect is powerful on its own. Avoid stacking too many additional complex LoRAs or prompts that might conflict with the fractal geometry.

3. WARNING! This LoRA is NOT compatible with the refusal bypass techniques like TextFusion and Conditioning Rebalance and they will break / weaken the fractalic effect. Maybe they can work at lower lora strengths / weights, but I didn't experiments with these parameters yet. Before making any complaints, please make sure they are turned off.

1. Update: it is compatible with TextFusion, but use with caution still!

4. Experiment with Scale: The model's ability to handle both large-scale environments and tiny, self-similar details is impressive. Try prompts that play with these scale shifts. Higher resolution is appreciated: the training is 2K, so 2-3 megapixels is a must for the best effect.

5. The LoRA can add repeats of the objects, spirals or fabric at the borders of the images. If undesirable, prompt them out with negative prompts and adjustments to the positive prompt.