August 12, 2026
Creating your first AI image: the guide for absolute beginners
From signing up to the downloaded file, the whole path to a successful first AI image on Imaginode. Nothing to install, about a cent to try, and a simple method to avoid the mistakes that make beginners quit.
Why this is much simpler than you think
Plenty of people put off their first AI image because they imagine an obstacle course: software to install, an €800 graphics card, command lines to memorize. None of that exists on Imaginode. Everything happens in your browser, generation runs 100% on the server side, and your old laptop from 2019 works perfectly well, since it never has to compute anything itself.
The other blocker is the imagined price. People assume a mandatory subscription and a bill that balloons at the end of the month. In reality, a first image on the Flux Schnell model costs 1 credit, roughly one euro cent. You can fail ten attempts in a row for the price of a stick of gum. That number changes everything: it turns experimenting into a reflex rather than a decision you have to weigh.
This guide takes you from zero. By the end, you will have created an account, opened a project, written a decent prompt, generated several versions of an image, and downloaded the one you like. Count on twenty minutes, coffee breaks included, and no technical background required at any point along the way.
What you actually need before you start
The list is short. A recent browser, an email address, and that is about it. There is no minimum spec to check, no graphics driver to update, no 40 GB folder to download. Chrome, Firefox, Safari, or Edge all work fine, on Windows as well as Mac or Linux. If your machine can play a video online, it can run Imaginode.
You do not even need a computer. Imaginode works in your phone's browser, with a fully touch-enabled canvas: one-finger panning, pinch to zoom. For a first contact, a computer screen is still more comfortable, but know that nothing forces you to use one. A generation launched from your phone keeps running even with the screen locked.
As for language, the interface is available in five languages, French included. Prompts, on the other hand, give better results in English, but as you will see further down, one button solves that problem for 1 credit. Do not let your level of English stop you before you have even started.
Creating an account and opening your first project
Signing up on imaginode.ai takes a minute. Once logged in, you land on your projects. Create a new one and name it whatever you like: "first attempts" works just fine. A project is a canvas, a large work surface where you will place your elements. Everything you do there is saved automatically, so you lose nothing by closing the tab.
The empty canvas can be intimidating for a second. That is normal. Unlike classic AI tools where you type text into a single box, Imaginode works with nodes, blocks you place on the canvas and connect to each other with cables. For a first image, you will only need a single block. The wiring logic comes later, once you actually have a use for it.
If the canvas loses you, an assistant bubble lives in the bottom right corner of the screen. It answers questions while seeing your open canvas, and each message costs 1 credit. Think of it as your lifeline, and remember it exists before you close the tab in frustration.
The Image node, your first tool
Add an Image node to the canvas. The block that appears contains the essentials: a prompt field where you describe the image you want, a model selector, and a Generate button. It is deliberately stripped down. No twenty obscure settings, no "sampler" values to understand before you can produce anything at all.
Take a good look at the Generate button: the exact cost in credits is displayed on it before you click. That detail seems minor, but it is actually the best guardrail a beginner can have. You always know what you are about to spend, down to the credit, before spending anything. No nasty surprise at the end of the month, ever.
The node also remembers what it produces: its last 12 generations stay available in a built-in history. You can therefore compare your attempts side by side without saving anything manually. We will come back to this, because that history genuinely changes the way you work.
Writing a first prompt that holds up
The prompt is the text description of the image you want. The classic first-day mistake: writing "a cat" and being surprised by the generic result. The model does not invent what you leave unsaid, it fills in the blanks with its habits. A cat, sure, but which cat, where, under what light, photographed how?
A simple structure already gives good results: the subject, what it is doing, the setting, the light, the style. For example: "a ginger cat asleep on a windowsill, Paris apartment, soft late-afternoon light, realistic photography". Thirty extra seconds of thought, and the image goes from bland to something you would actually show someone.
Do not aim for perfection on the first try. Nobody, not even the creators who make a living with these tools, gets the final image on the first attempt. The initial prompt sets a direction, and the following versions correct course. We have devoted an entire article to prompt writing, by the way. Save it for the next step.
Choosing a model without losing your whole evening
Imaginode gives you access to 47 models with the same account and the same credits, including about ten for images. Faced with that list, a beginner has two possible reactions: paralysis, or picking the most expensive one "just to be safe". Both are mistakes. To start out, there is only one name to remember: Flux Schnell, at 1 credit per image.
Flux Schnell is not the most beautiful model out there, and it does not pretend to be. It is fast and nearly free, which makes it the ideal iteration tool. Between the two extremes, Flux 2 Klein and Imagen 4 Fast run at 3 credits with a very good quality-to-price ratio, and Flux 2 Pro tops out at 7 credits for the most polished rendering.
There are also specialists: Seedream 5 Lite at 5 credits excels at characters with references, and Flux Kontext at 6 credits is built for editing an existing image. Ignore them for now. Your first goal is to understand the prompt, generate, correct loop. The right model for learning is the cheapest one.
Credits, explained without the fog
One credit is worth about €0.01. That is the only conversion you need to memorize. A Flux Schnell image therefore costs one cent, a Flux 2 Pro image seven cents. With the Starter plan at €13 per month before tax, you receive 900 credits, enough to generate 900 drafts or 128 images at maximum quality. In practice, you will do a mix of both.
If subscriptions put you off, one-time top-ups with no commitment exist: 1,000 credits for €15, up to 10,000 for €135. The public code SIMPLE15 takes 15% off along the way. And every credit movement, purchases as well as spending, shows up in a complete history on your profile. You always know where every cent goes.
One point that reassures cautious people: if a generation fails for a technical reason on the model provider's side, the credits come back to your account automatically within seconds. You never pay for an image that does not exist. The only credits truly spent are for delivered images, including the failed ones, but that is part of the game.
BalanceClicking Generate, and what happens next
Prompt written, model chosen, cost checked on the button. Click. The generation runs on the servers and the image appears in the node after a few seconds, the exact wait depending on the model. With Flux Schnell, it is nearly instant. There is nothing to monitor, nothing to download; the image lives in the node.
First reflex to build: look at the whole image before judging it. Zoom in on the hands if your subject is human, and check any text if there is some. Image models have improved enormously, but extra fingers and gibberish signage remain their known weaknesses. Better to spot them now than after sending the image to a client.
Second reflex: delete nothing. Every generation stays in the node's history, and an image judged a failure at 2 pm sometimes turns out to be the best one at 6 pm, once you compare it with the ones that came after. Snap judgment is a poor advisor; the history is there for exactly that reason.
The golden rule: iterate at 1 credit, finalize at 7
Here is the method that separates beginners who improve from beginners who give up. Always work in two phases. First, explore with Flux Schnell at 1 credit: test your prompt, fix it, change the light, try another framing. Ten attempts cost you ten cents. This is your sketchbook, and you should treat it exactly like one.
Then, and only once the prompt reliably produces the composition you are after, switch to a high-end model like Flux 2 Pro at 7 credits for the final version. Same prompt, better rendering. You get maximum quality while having paid full price only once or twice in the whole process.
Run the numbers in reverse to measure the savings. Ten iterations directly at 7 credits: 70 credits, about 70 cents. Nine iterations at 1 credit plus one final at 7: 16 credits. More than four times cheaper, for an identical end result. Over a month of regular practice, this one simple habit is worth several euros.
The mistakes every beginner makes
The first: the river-length prompt right from the start. Forty words piled up, half of them contradicting each other, and no way to know which word produced which effect. Start short, with one solid sentence, then add one element per generation. You will always know what just changed, and why.
The second: regenerating the same prompt in a loop and hoping for a miracle. If three consecutive generations produce the same flaw, luck will not fix it; the prompt is what needs to change. Every click without a change is a credit spent testing your luck rather than your idea. One cent, sure, but above all wasted time.
The third: expecting the model to do what it cannot. A character who keeps the same face from one image to the next, for example, is not something a longer prompt will get you. Without reference images, the face changes with every generation; that is structural. Imaginode has a Reference node for exactly this, but that is a topic for your second week.
Per-node history, your safety net
Every Image node keeps its last 12 generations. Concretely, you can chain attempts without any file management: no "versions" folder on your desktop, no backup screenshots. Everything stays attached to the node that produced it, along with the prompt that went with it.
That history changes the psychology of iteration. When going back costs nothing, you dare more. You can try a radically different prompt, compare, and return to the previous version in two clicks if the experiment fails. Add the canvas undo-redo and the automatic project save on top, and it is nearly impossible to lose work.
One honest limitation all the same: 12 generations, not 200. During a long session around the same node, the oldest attempts eventually fall out of the history. When a version feels precious, download it or duplicate the node. Thirty seconds of caution for the images that really matter.
Download it, then show it
Happy with your image? Download it from the node, and the file lands in your downloads folder like any other image from the web. It is yours, ready for a post, a mockup, a presentation. No convoluted export step, no watermark you have to pay to make disappear.
A piece of advice from experience: show this first image to someone. Not for the glory, for the outside eye. You develop a tolerance for your own AI images very quickly, and a fresh pair of eyes spots in three seconds the detail that is off, the strange hand or the inconsistent shadow you had stopped seeing. It is the cheapest quality control in the world.
Also keep the winning prompt somewhere, for example on a sticky note placed directly on the canvas, a feature designed for exactly that. A prompt that worked well is capital: you will recycle it, adapt it into variations, and a month from now it will serve as the base for images you cannot even imagine yet.
The magic wand when your English gets stuck
Let's talk about the sore spot. Image models will take prompts in any language, but they were trained mostly on English descriptions, and a precise English prompt gives visibly better results. Frustrating when English is not your strong suit, or when you simply have no idea how photographic English says "backlit".
The magic wand, present on the prompt field of every Image node, solves the problem for 1 credit. You write your idea simply, in your own words, and it rewrites the whole thing in rich, structured English, with the technical vocabulary the models expect. Your "cat on a window at sunset" becomes a real photographic direction.
The right way to use it: write your intention without censoring yourself, run the wand, read the result. Even without speaking English fluently, you will quickly recognize the recurring terms, golden hour, shallow depth of field, and you will pick up the trade's vocabulary while working. At one cent per lesson, it is the cheapest photography course on the market.
A complete example, stopwatch in hand
Let's run through a real case from start to finish. Goal: an illustration image for the home page of a burger food truck. First prompt, deliberately simple: "a generous burger on a wooden board, food truck in the background, photography". Generated on Flux Schnell, 1 credit, result in a few seconds. Decent, but the light is flat and the background is muddled.
Second attempt: adding "golden late-day light, steam rising from the patty". Clearly better. Third attempt: "close-up, shallow depth of field" to separate the burger from the background. Fourth: the bun looks plasticky, so we specify "artisanal brioche bun, sesame seeds". At this point, 4 credits spent, and the composition is locked in.
A pass with the magic wand to enrich everything in English, 1 credit, then the final generation on Flux 2 Pro, 7 credits. Total for the whole operation: 12 credits, about twelve cents, in a quarter of an hour. An equivalent burger photo from a photographer bills between €150 and €400. The comparison is not entirely fair, a real photographer shoots your real burger, but for a mockup or a concept test, the math takes about two seconds.
Above all, remember the mechanics: each attempt fixed only one thing. Light, then framing, then texture. It looks slow on paper, but it is actually the shortest path, because at no point do you lose track of what is working.
What comes next: video and workflows
Once the first image is under your belt, two doors open. The first is video. A Video node can take your image as a starting point and animate it, from about 26 credits for 5 seconds with Kling 2.5 Turbo. Fair warning: AI video still fails about one generation in three or four, between questionable physics and details that melt. It also sometimes pulls off shots no budget would ever have bought you.
The second door is the canvas itself. Connect a Text node that writes descriptions to your Image node, add a Style node to enforce an art direction, and you have a small workshop that produces consistent series of images. This is where Imaginode really stands apart from single-box generators: the ready-made templates give you complete setups to study.
Concrete next step: generate five variations of today's image, changing only one word of the prompt each time. Total cost, five cents. It is the most effective exercise for understanding what actually carries weight in a prompt, and it leads you straight into our guide on writing prompts that work.