What Krea’s One Login Actually Replaces

Chapter 1

One Subscription Replaces the Six Tabs You Had Open

A freelance social media manager keeps six tabs open before lunch. One AI image generator for the moodboard. Another for upscaling the client's logo. A third for turning a still photo into a short teaser. A fourth for removing a distracting background. A fifth waiting on an API key that hasn't been approved yet. A sixth just for comparing which of the first four actually gave a usable result. Five logins, five separate bills, and by the time all five have loaded, the actual work still hasn't started.

Krea's whole premise is that those six tabs collapse into one. The same account that generates an image can edit it, upscale it, turn it into a short video, train a look that stays consistent across a whole campaign, and, for anyone building a product feature on top of it, call the same models through an API. The models themselves aren't Krea's own inventions in most cases – Flux, Imagen, Nano Banana, Veo, Kling, Sora and dozens more all sit behind the same prompt box, so switching between them is a dropdown, not a new signup.

The site states it is used by tens of millions of people across more than 190 countries, spanning individual creators up to companies large enough to run enterprise workflows. That range matters for how the rest of this book is written: a hobbyist testing a free tier and a brand team training a custom model are both reading the same pricing page, and the page has to serve both.

What actually holds all of it together is a single currency: compute units. Every generation – an image, an edit, an upscale, a video clip – costs a number of units, and which plan someone is on decides how many units they get, how many jobs can run at once, and which model families they're even allowed to open. A free account gets a small daily allowance and no video generation at all. Paid tiers open video access, faster concurrency, and higher-resolution upscaling in stages, not all at once.

That's the real decision a new user is making – not which AI tool is best in the abstract, but which plan actually matches how often they'll use this. The next chapter breaks the tiers apart, because the difference between them is not only price. It's what each one actually lets a reader do.

What you can actually do here

Krea bundles a lot behind one login. This map groups what it actually does by the kind of work each group solves, so the rest of the book can go deeper on the ones worth it.

Generating and editing images

Making video from nothing

Staying consistent

Automating and building

Fixing and finishing

Chapter 2

What a Compute Unit Buys You, Plan by Plan

The five tiers on Krea's pricing page look, at a glance, like the same list every subscription service uses: Free, Basic, Pro, Max, and then Business and Enterprise for teams. What separates them is not simply a growing pile of units. Each step up unlocks a different category of work entirely, and reading the tiers as 'more credits' misses the part that actually changes what someone can build.

Free gives a small daily unit allowance and access to every image model, but video generation is switched off completely, LoRA training is capped low, and only one job can run at a time. That's enough for someone testing whether the tool fits their workflow at all. It is not enough for anyone producing on a schedule.

Basic is the first tier where video generation appears, though only on a selected set of models rather than the full roster, and image concurrency moves from one job to several. Pro is where every video model becomes available and concurrency opens further, which is why the pricing page marks it as the plan most people land on. Max removes the concurrency ceiling entirely and raises the per-LoRA image cap from the tens into the thousands, which only matters to someone training many custom looks at once.

Business and Enterprise exist for a different problem than personal use: seats, roles, per-member spend limits, and a Terms of Service built for an organization rather than an individual. A team choosing between Business and Enterprise is choosing between a self-serve workspace and a negotiated one with dedicated support – the features overlap heavily, but the second is priced and configured directly with Krea rather than picked off a page.

One detail belongs here because it's easy to miss until the bill arrives: compute purchased outside a plan, in a one-time pack, is not permanent. It's added instantly and stays usable for ninety days, then it's gone whether it was spent or not. Anyone buying a pack to cover a busy month should plan to use it inside that season, not bank it for later.

None of this settles which plan is right for any one reader – that's a decision the closing chapter comes back to, once the actual workflows in between have been walked through. For now, the plan only sets the ceiling. What happens inside that ceiling is the more interesting part.

Chapter 3

📰 Recommended for You

How Canva Turns an Indian Occasion Into a Printed Invitation

Read More →

What Every Pixlr Plan Actually Buys You

Read More →

From a Blank Prompt to a Finished Image

A teacher preparing a set of slide illustrations has a rough idea in her head – a friendly, hand-drawn style, no stock-photo stiffness – and no budget for an illustrator. She types a description into Krea's Image tool, picks a model from the dropdown, and within seconds has four visual interpretations of the same sentence. None of them are exactly right yet, but one is close enough to refine, which is a faster starting point than a blank page ever was.

That first generation is the shallow end of what the Image tool does. A prompt alone drives text-to-image generation; uploading a reference photo instead – or alongside the prompt – drives reference-image generation, which is how someone guides subject, composition, or product detail more precisely than words allow. The same workspace runs both, so a project can start from a sentence and finish from a photo without switching tools.

Editing lives one tab over, on the same account and the same credits. Instead of a mask-and-brush workflow, the Image Editor takes a written instruction – remove this object, change this background, extend this frame – and produces the change directly. Object removal, background replacement, image expansion for a new aspect ratio, and combining multiple reference images into one composition are all the same underlying move: describe the change in plain language, generate, compare, refine.

Different edit types call on different underlying models. OpenAI's model, Nano Banana, Qwen, Flux Kontext, Flux, and Ideogram are all available inside the same editor, and the site's own examples show them applied to product placement, style filters, and detail-level fixes rather than full regenerations. Choosing the right one for the job matters less than testing two or three on the same edit, which the workspace makes cheap to do.

The one place this needs a second look before publishing anything commercial: precise product text, logos, and packaging details can drift slightly during a generative edit, even when the rest of the photo holds up. A generated result is a strong starting point for a campaign visual – it is not yet a proof for print without someone checking the small print by eye.

Chapter 4

Turning a Photo or an Idea into Video

An illustrator has spent a week finishing a character design and has exactly zero animation experience. Hiring a studio for a six-second loop isn't worth it for a single social post, and every tutorial she finds assumes software she doesn't own. What she actually needs is one clip of the character blinking and turning its head – not a production, just proof of life.

Krea's Video tool works from three starting points: a text prompt alone, an uploaded image as the starting frame, or an existing clip that gets extended into something longer. For the illustrator's case, the second path is the one that matters – the character illustration becomes the start frame, and a written description of the motion ('the character blinks slowly, then turns its head toward camera') tells the model what to do with it.

Model choice changes the result more than the prompt does, in this case. The documentation is direct that Hailuo 2.3 is the platform's strongest option for 2D animation and cartoon styles specifically, because it keeps a character's appearance stable across frames in a way some of the larger, more expensive video models don't prioritize. It also runs on noticeably fewer compute units than most of the higher-tier models – a real cost, but a smaller one.

Every clip, on any model, tops out at twelve seconds before it needs to be extended rather than regenerated longer in one pass. Extending works by hovering over a finished clip, clicking Extend, and letting the model continue from the final frame with the same or an adjusted prompt. Longer sequences are built this way – clip, extend, extend again – rather than requested whole.

Some models are built for speed over polish, some for realism over speed, and a few – Veo 3.1 among them – generate synchronized audio alongside the picture rather than leaving sound for later. Running the same prompt across two or three models before committing to one is often faster than trying to guess which will fit a particular shot from the name alone.

Chapter 5

Fixing a Blurry Photo Without Losing What Made It Real

Two tools on Krea open on nearly identical upload screens, and a first-time visitor could be forgiven for treating them as the same feature. They aren't. The Upscaler exists to raise resolution – take a smaller image and make it larger without losing sharpness. The Enhancer exists to reconstruct detail that compression or blur has damaged. Using the wrong one first wastes a generation on a problem it wasn't built to solve.

The Upscaler's job is comparatively simple: 2x, 4x, 8x, or 16x resolution increases, with a Topaz-powered mode that reaches as high as 22K for print-scale output. It works on video as well as still images, and it batches – multiple photos can be queued and processed together rather than one at a time, which matters for anyone upscaling a full product catalog rather than a single hero shot.

The Enhancer goes further than resolution alone. It reconstructs texture, sharpens fine detail, and reduces noise and compression artifacts – the kind of cleanup an old scanned photo or a heavily compressed social upload actually needs before it can be printed or reused. Its own documentation is honest about where that stops: the model can make an intelligent guess about missing detail, but it cannot recover information a source image never captured in the first place. An extremely blurry, heavily pixelated, or motion-blurred photo has a real ceiling no amount of AI reconstruction crosses.

A practical order of operations follows from that limit. If the goal is a bigger file at the same quality, start with the Upscaler. If the goal is a better-looking file from a damaged source, start with the Enhancer, and treat the Upscaler as the second pass once the detail has already been recovered.

Anyone working past a certain output size – the higher upscaling tiers, batches of product photos, or exported video at 4K – will notice how quickly the files add up. Storage and transfer speed become the quiet cost of doing this well, long after the compute-unit bill has already been paid.

Chapter 6

Keeping One Face, One Brand, One Mood Across Hundreds of Generations

Three tools on Krea all solve some version of the same problem – keeping a look the same across many generations – and they aren't interchangeable. Style References transfer the visual character of up to four uploaded images onto a new prompt: focused, fast, and tightly controlled. LoRA training goes further, fine-tuning a small custom model on a product, character, or brand so that identity persists automatically, generation after generation, without re-uploading references each time. Mood boards do something different from both.

A mood board isn't capped at four images the way a style reference is – a user can upload as many references as they want, and the system runs a more involved process underneath: custom models read style, concepts, expressions, and overall mood from the whole set, not just color and texture. Before a board can be used, it has to be analyzed once, which produces three things – a taste profile describing what the system found, a set of keyword tags applied under the hood, and a list of things the system will actively avoid.

What that analysis actually changes is easy to underestimate until it's demonstrated. Krea's own documentation shows the same simple prompt – 'a frog' – run through three different mood boards and returning three visibly different worlds: a starry, wide-eyed board produces a frog in a starry scene with a similarly wide-eyed expression; a different board pushes the same prompt somewhere else entirely. Style references would have matched a palette. The mood board matched a whole aesthetic sensibility.

LoRA training solves a narrower, more durable version of the same need – a product or character that needs to look identical across an entire campcampaign's worth of images, not just one session. Training image caps scale hard with plan tier: the lower tiers allow a modest number of training images, while Max and Business allow thousands, and Enterprise negotiates a custom cap. A brand testing whether custom training is worth it should start on whichever plan already covers the training-image count their product actually needs.

None of these three tools replaces the others. A single project, in practice, often uses a mood board to set the overall direction, a style reference to lock one specific look within it, and LoRA training only once a product or character needs to reappear identically across dozens of future pieces.

Chapter 7

Building a Workflow Once, Running It Forever

By the third week of a weekly content cycle, most of the actual work isn't creative anymore – it's repetition. The same steps, in the same order, for a slightly different product each time: generate a base image, upscale it, apply a brand style, export. Doing that by hand every week is the kind of task that quietly eats a whole afternoon without anyone noticing where the time went.

Nodes is Krea's answer to that repetition. It's a visual, node-based workflow builder that chains tools together so the output of one step feeds directly into the next – a generation node connecting into an upscale node connecting into an export step, run as one sequence instead of four separate manual jobs. Multiple models can also be run in parallel inside the same workflow, which turns 'which model looks best for this' from a guessing game into a side-by-side comparison built once and reused.

Once a flow works, it doesn't need to be rebuilt. It saves as a reusable workflow that runs again on the next batch without anyone re-wiring the connections, and it can be shared as a template with teammates – a detail the feature page mentions only briefly, in passing, though it's the part that actually matters for an agency onboarding a new team member onto an existing pipeline rather than teaching them the tool from scratch.

App Builder and Nodes Agent sit one layer above the basic workflow builder, aimed at teams that want to package a node workflow as something closer to a small internal tool, with an agent able to operate parts of the chain rather than requiring a manual trigger at every step. Both are newer additions and worth testing directly rather than assuming their scope from the name alone.

The honest limit here is a familiar one for any automation tool: a workflow only stays reliable as long as every model inside it keeps behaving the way it did when the workflow was built. If a chained model changes its output format after an update, the workflow needs a human who understands the chain to notice and fix it – Nodes removes the manual repetition, not the need to understand what it's automating.

Chapter 8

Where the App Credits Stop and the Dollar Bill Begins

A developer on a Max plan builds a prototype that calls Krea's image generation from their own product, tests it a dozen times against their own compute allowance, and is surprised a week later to find the API charging separately, against a balance they never funded. The confusion is understandable and common enough that Krea's own API FAQ addresses it directly: an app subscription – Free, Basic, Pro, Max, whatever tier – covers the Krea web app only. API usage draws from a completely separate, prepaid dollar balance.

That separate balance is billed in a genuinely different way from app credits. Instead of compute units, API pricing is dollar-denominated per generation, tracked internally in microdollars for exact accounting. A Nano Banana 2 image costs a stated $0.06; a Nano Banana Pro image costs $0.15; a second of Veo 3 video costs $0.20; a Flux image costs $0.04. Every model on the API price list carries its own rate and its own typical generation time, and failed or cancelled jobs are explicitly not charged.

What that buys, in exchange for managing a second balance, is real: one REST API in front of more than forty image and video models, rather than forty separate vendor accounts and forty separate SDKs to learn. A single SDK call – shown on the API page in JavaScript, Python, Go, or plain cURL – returns a result the same way regardless of which underlying model answered it, and swapping models is a one-line change rather than a new integration.

The platform's own FAQ is refreshingly plain about which model to start with: it depends on the job. Most teams begin with Flux for speed and move to Imagen 4 or Nano Banana Pro once fidelity actually matters more than iteration speed. Webhooks remove the need to poll for a finished job, and every request goes through the same queued-processing-completed lifecycle regardless of the model behind it.

So the decision this whole book has been circling comes down to a short list of questions, not a verdict. If the goal is occasional, personal image generation, the Free plan answers it and nothing more is owed. If video matters at all, Basic is the floor and Pro is where every model opens up. If a brand needs one consistent look reused for months, LoRA training and mood boards are worth planning the plan tier around, not treating as an afterthought. And if the goal is building a product feature rather than making a picture, the API's dollar balance – not the app subscription – is the number to budget, and it is worth pricing out against the specific models a product will actually call before committing to any of it.

Questions readers actually ask

Contact / More useful information from RamthaMedia

Official source links:
Krea