Hey hey,

Krea dropped a new model last week called Krea 2, and as of yesterday it's live on every paid plan with an unlimited promo running on top of that.

Normally I'd let a launch like this sit for a bit before saying anything, but the claim they made… I had to test it.

Most image models are good at putting the right stuff in the picture but you usually don't get control over how it looks.

You get something clean, safe, a little samey, the thing people call the AI look.

Krea built Krea 2 around the second half of that problem.

So I spent a few hours using it and here's how it went. But before that, some catch up:

1. Manus Scheduled Tasks: Set the autonomous agent to run recurring jobs on a timer, like a daily news brief or weekly competitor research, with results sent to email, a file, or a connected app. A recent upgrade also lets you pick which agent runs each scheduled task, so you can control cost and reliability on high-volume workflows.

2. Polarity: Sandboxed eval infrastructure that runs each AI agent task inside an isolated Docker sandbox loaded with real backing services like Postgres, Redis and S3, then scores the run against behavioral rules.

3. LobeHub: A Chief Agent Operator: one workspace where you hire, schedule, and run a whole team of AI agents that operate around the clock and report back without you babysitting them. It grew out of the open-source LobeChat framework, so it spans providers like GPT, Claude, Gemini and DeepSeek, with shared context, scheduled tasks, and project organization built in.

Krea 2 is Krea's first image model built fully in-house, from scratch. Before this they mostly wired up other people's models.

It's quick. Generations land in around 15 seconds, and has been tuned for range. Photoreal works, but so do illustration, painterly stuff, manga, anime, even grainy VHS looks.

In most tools, style is a word you bury in your prompt and hope for with the likes of "Cinematic." "Watercolor." or "Editorial."

In Krea 2, style is a thing you hand it and then control.

You drop in a reference image, and it reads the visual language out of that image, colours, brushwork, tone, and carries it into whatever you generate.

You can stack more than one reference, and slide how strongly each one pulls, so a reference can be the whole look or just a nudge in a direction.

Moodboards are the same idea, scaled up.

One reference image can't always carry a vibe. So for the looser stuff, the aesthetics that live across five or six images instead of one, you build a board.

Curate the images, point Krea 2 at the board, and it generates inside that world.

You can also share a board with a public link, which is handy.

1. Pick the model

Open image mode, click the model name in the bottom left, choose Krea 2.

2. Run a plain prompt first without references

I started deliberately basic to see the spread. It doesn't hand you the same idea four times with tiny differences and the outputs diverge.

For early ideation that's useful, you see options you wouldn't have prompted your way to.

Prompt: a cup of coffee on a cafe table, morning light.

Output:

3. Add a style reference

Then I dropped in one reference image (a warm and grainy faded film still) and ran the same prompt again.

The shift was obvious.

The same coffee cup with a completely different skin on it, and it held the reference instead of drifting back to a generic finish.

Prompt: a cup of coffee on a cafe table, morning light.

Output:

4. Build a small moodboard

Next I pulled together a handful of images into a board and generated off that instead.

This is where it gets useful for brand work. The output sat inside the mood without me describing the mood in words at all.

It came back with chunky handmade clay mugs, speckled stoneware, linen folded next to the cup, a rust and cream palette.

I never typed the words terracotta, linen, or handmade anywhere.

Prompt: a cup of coffee on a cafe table, morning light.

Settings: Moodboard selected

Output:

5. Dial the strength

I turned style strength up, creativity up, and randomness up, and generated.

This one was my favourite set of the lot because of the variety / randomness as it introduced a human to the picture.

And it felt art-directed, like someone with an eye had been in the room.

Plenty of it misses, but the ones that landed look very deliberate rather than hedged.

Prompt: a cup of coffee on a cafe table, morning light.

Output:

The whole loop took me 30 minutes including generating the references and everything.

A few things I'd reach for it for:

The common thread: you already have a sense of how it should look. That's when Krea 2 earns its place.

I love how little prompting I had to do once I started feeding it references. Finally, less typing and more vibe surfing.

If you need precise instructions followed to the letter, or a lot of clean readable text inside the image, Nano Banana Pro or GPT Image 2 are still the better call.

There's a benchmark from Contra Labs that lines up with what I felt: Krea 2 came in second for style transfer, sitting only 0.14 points behind GPT Image 2 on style fidelity, while being weaker on the prompt-led, literal stuff.

That's roughly the experience.

On cost: there's a free tier with daily credits if you just want to poke at it, but Krea 2 itself sits on the paid plans, and there's an unlimited promo running across paid plans right now.

Most image tools reward people who are good at writing prompts while Krea 2 rewards people who already have taste and just need a faster way to get it out.

If that's you, this one's worth the weekend experiment.

Until next time,
Vaibhav 🤝🏻

If you read till here, you might find this interesting

#AD 1

The DTC beauty category is crowded. To break through, Jennifer Aniston’s brand LolaVie, worked with Roku Ads Manager to easily set up, test, and optimize CTV ad creatives. The campaign helped drive a big lift in sales and customer growth, helping LolaVie break through in the crowded beauty category.

Learn more

#AD 2

Power users figured this out early: speaking a prompt gives you 10x more context in half the time. You include the edge cases, the examples, the tone you want — because talking is fast enough that you don't skip them.

Wispr Flow captures everything you say and turns it into clean, structured text for any AI tool. Speak messy. Get polished input. Paste into ChatGPT, Claude, Cursor, or wherever you work.

89% of messages sent with zero edits. 4x faster than typing. Works system-wide on Mac, Windows, and iPhone.

Start flowing free