fal-Krea-2-Style-LoRAs

<picture> <source media="(prefers-color-scheme: dark)" srcset="https://huggingface.co/ilkerzgi/fal-Krea-2-Style-LoRAs/resolve/main/fal-dark.png"> <img src="https://huggingface.co/ilkerzgi/fal-Krea-2-Style-LoRAs/resolve/main/fal-light.png" alt="fal" height="22"> </picture>

fal · Krea 2 Style LoRAs

1503 style LoRAs for Krea 2, trained on fal. Each LoRA has its own repo with the weights (the original fal/diffusers safetensors plus a ComfyUI-compatible version), a model card, and copy-paste fal usage. Browse the full index below.

<video controls autoplay loop muted playsinline width="100%" src="https://huggingface.co/ilkerzgi/fal-Krea-2-Style-LoRAs/resolve/main/reel-999.mp4"></video>

Author's note

All of these LoRAs were trained on styles I picked over the years: styles I saved because I loved them, or favorites pulled from images I generated myself. Some used around 5 images, others around 10.

I'm obviously not claiming you can train a flawless, extraordinary LoRA from such a tiny dataset. But if you look at the examples below, getting results this good, especially for style, with this few steps was very hard on earlier models. While exploring Krea 2, I trained a few LoRAs just to expand the styles I already had; once I saw they were getting genuinely good feedback, I trained my whole archive the same way, and the next day, once they were all ready, I released them open-source. So it really was a completely spontaneous thing.

Feel free to take any of these style LoRAs you like and push them further with a larger dataset and more steps. That is exactly what the open-source community is for.

If you find any LoRAs that are broken, low quality, or have broken weight links, you can flag them in that LoRA's repository. I will fix and, if needed, retrain it. And if you have trained other Krea 2 LoRAs, or retrained any of these, share them and I will add them here.

Low-step training comparison on other base models

Bold Stripe Pattern Render, 5 training images, 100 steps on each model. Reference style:

How to use

Pick a style below and copy its safetensors link, then paste that link into the LoRA field on fal-ai/krea-2/turbo/lora and add the style's trigger to the end of your prompt (recommended LoRA scale 1.0 to 1.25). To use a style in ComfyUI instead, download its comfy safetensors link and load that. Tip: shorter prompts give better results, so keep them concise, describe the subject simply, and always put the trigger at the end. Long, overloaded prompts tend to wash the style out.

Styles by category

3D & Render (97)

Cinematic (101)

Drawing (98)

| Preview | Style | Trigger | safetensors | comfy |

The model card is truncated. Read the rest on Hugging Face.