introducing LoRAs for Krea 2 (beta). our most powerful fine-tuning system to date; now you can train Krea 2 on a your own specific style, object, or character with incredible precision. learn how it works 👇

Krea 2 LoRAs are now available for Max and Business subscribers as we test them in Beta. LoRAs are not new to Krea, we had them available for a while with models like Flux, Qwen, or Wan. though, with Krea 2 the quality and precision of LoRAs is way higher than anything we had until now.

to start, head to select "Train new Lora", and then press "Next step".

then, select a set of images and add them into the trainer. on the right sidebar, make sure that "Krea 2 Medium" or "Krea 2 Large" are selected. you can change settings like "Steps", which will determine how strong the training is.

once the LoRA is trained, you can use it on the Image tool. you can control its strength and combine multiple LoRAs together.

this was literally the only feedback I mentioned when I asked for some, hugeee

🙌🙌🙌

again testing... :D

❤️❤️❤️

🔥🔥🔥

Huge 🔥🔥👌

wow looks insane. can’t wait to try this

amazing 👌

Finally I can stop describing my style in prompts and just train it once

Custom AI styles are getting insanely good 🎨🔥 Train Krea 2 on your own characters, objects, or visual style with crazy precision. The era of truly personalized AI visuals is here 🚀

Exactly what I was waiting for...

Oooh! This is what I want! Look forward to custom LoRas!!! Yes!

fine-tuning can be game-changing, but how's it handle edge cases like niche styles?

@viccpoes excited to play with this

first thought i had: what's the difference between moodboarding and loras? moodboarding guides the model toward a style through prompts, it does its best to match lora training makes the model internalize the style completely, so whatever you generate comes out in that aesthetic by default, very cool update!

Simply Amazing 👏🏻

@viccpoes Train your perfect style

Many popular front-facing AI image platforms still don’t offer users direct LoRA training. Tools like DALL·E (via ChatGPT) and Google’s Nano Banana keep everything optimized behind the scenes. You can usually get decent consistency with reference images or in-context editing, but full custom LoRA training on your own styles, characters, or objects isn’t available.Krea is changing that with their new LoRA feature for Krea 2. Creators can now upload their own image sets, choose between Krea 2 Medium or Large, and adjust training strength to build truly personalized models directly on the platform. A meaningful step toward giving users real control over consistent styles and characters

screenshot saved

This is a great upgrade. Training custom styles and characters with more precision will be really useful for creators.

the real unlock is not just more precise training data but collapsing the time between data gathering and application. no one talks about the lost hours in data prep, but loras changes that.

this is the Krea feature we’ve wanted for a while film work breaks when character, costume, texture, and visual taste have to hold across a real sequence curious if Krea 2 LoRAs are stronger for identity, style, or both worth exploring a small creator partner pilot?

Awesome news

The next big unlock would be: cross-media style consistency. Imagine one LoRA controlling: → images → video → UI design → animations → 3D assets So creators maintain one recognizable identity everywhere automatically.

Please open it up to Pro Plan users, would love to try🙏❤️

Custom LoRAs unlock serious control

Krea 2 LoRAs give creators precise style control without the hassle

🔥🔥