How to use Krea AI Realtime generation

Krea AI Realtime updates an output as you draw or change its inputs: open Realtime, enter a prompt, and adjust the input canvas while watching the result. The Realtime guide↗ (opens in a new tab) places the input on the left and output on the right. Save a useful image along with the inputs that led to it. fenbs can keep those findings so your next assistant knows which visual direction worked and what you still need to test.

Use the live canvas to answer a small question, such as whether a large central shape or an off-center arrangement suits a poster. Set one goal before touching the controls. A stream of interesting pictures is easy to produce; a useful result needs a reason to keep it.

Where do you find the live canvas controls?

Open Realtime from the Krea dashboard or at www.krea.ai/realtime. The selector below the prompt offers Draw, Web Camera, Screen Record, and Text Only. Image/Video and aspect ratio controls sit below the prompt too, while Model is at the bottom left. Those locations are documented in the Realtime guide.

Begin with Draw for a simple layout test. Web Camera uses a live camera feed; Screen Record uses a browser tab, so do not assume it observes every desktop application. Choose the input that matches the experiment rather than changing several sources at once.

Check available models in your account before committing to a workflow. The Krea pricing page↗ (opens in a new tab) compares access, but the presence of a model in a guide does not mean every account can use it. This walkthrough follows the browser interface rather than promising identical controls in every mobile app.

How do you run your first drawing test?

1. Open Realtime and choose Draw below the prompt. Choose Image for a still-image experiment and an aspect ratio suited to the intended layout.
2. Use Model at the bottom left to choose an available model. Write down its name so you know what generated the output.
3. Enter a short prompt describing the scene and visual treatment. For example: “Editorial illustration of a yellow vase on a navy table, simple painted shapes, generous empty space above.”
4. Draw a rough arrangement on the left input canvas. Watch the right output as you change the arrangement.
5. Stop when you have something worth comparing. Note the prompt, input shapes, model, aspect ratio, and AI strength before making the next change.

You can also use Examples to load a preset canvas and prompt, as the guide describes. Treat a preset as a starting example: change one element and observe what happens. If you replace both the entire drawing and the prompt, you lose the chance to learn which input mattered.

The vase example is a suggested experiment, not a tested output. For the wider process of moving from an idea to reviewed artwork, read the AI workflow for designers.

What should you change when the output drifts?

AI strength adjusts how closely the output follows the input. The Realtime guide specifically recommends lowering it when you want camera output to remain closer to the feed. For a canvas experiment, compare a small adjustment against the same prompt and drawing, then decide whether the relationship is more useful.

If the composition is wrong, first inspect the layout you supplied. If the composition works but the treatment is wrong, change the descriptive words while holding the drawing steady. This is a review method, not a guarantee about how a particular model will interpret a mark.

• Wrong arrangement: compare a revised input shape position with the previous arrangement.
• Too much departure from the source: test AI strength while keeping the other inputs steady.
• Useful direction, unwanted variation: keep the accepted output before pressing Seed for a fresh result.
• Hard to judge improvement: compare both versions at the size and crop you actually need.

Seed creates a new variation; it is not a documented promise that you can reconstruct an identical image in another session. Keep the evidence you have, rather than treating the button name as a complete reproduction method.

How do you tell whether an experiment worked?

Write down the requirement before deciding which output is your favorite. For the vase example, you might require a recognizable vase, a clear separation from the background, and enough room above for a headline. An attractive image can still fail the intended layout.

Compare an accepted version with one rejected version and explain the difference in a sentence. “Version B leaves room for the heading; version A fills the entire frame” is useful evidence. “B looks better” leaves the next person guessing what must stay.

The design review guide covers checking against requirements, and our guide for graphic designers looks at the broader workflow. Save your concrete finding in fenbs so another assistant can help without reopening the same comparison.

What does Download keep, and what should you record?

At the bottom of the output panel, Download saves the image to your device. Upscale sends the result through Krea’s upscaling pipeline. If the image needs that separate treatment, use the Krea enhancement walkthrough rather than assuming the live preview is the final production asset.

A downloaded image is not a documented bundle containing the editable canvas, prompt, seed, and every setting. Make a separate record of the prompt, input source, model, aspect ratio, and AI strength used. Keep the original inputs and identify any settings you did not record instead of filling the gaps from memory.

Krea also documents its own durable context and saved workflows in Krea Agent’s context library and skills↗ (opens in a new tab). Those are useful when you continue inside Krea Agent; they do not establish a complete Realtime session-export feature. Choose where to keep your record based on who needs to use it next.

How to keep your Realtime findings with fenbs

Keep the useful work so you can carry on from exactly where you left off, on any device and in any other connected AI. A short record of the accepted direction helps more than a folder of pictures with no explanation.

1. Create a free account.
2. Save your accepted prompt, the input you used, the model name, and the settings you recorded.
3. Add a link to the downloaded image. Write why you kept it and what still needs checking.
4. Connect a supported assistant from Integrations and ask it to read this work before helping with the next version.
5. Save what you tried and the next small test before you stop. Use that same record when you return.

fenbs does not import your Realtime canvas or generate Krea images. Krea’s own tools may be enough when all your work stays there. Use fenbs when findings and next steps need to travel between assistants; explore other practical workflows in AI at work.