Krea AI Prompts

Krea AI was founded by Diego Rodriguez and Victor Perez and is headquartered in San Francisco. TechCrunch reported a $5M seed round led by Andreessen Horowitz in March 2024, and Krea's positioning then and now is that it is a real-time creative tool rather than another text-to-image box. The product has since grown into an image, video, training, and depth-aware staging platform with Krea 1, the in-house image model, sitting alongside outside models the team curates for specific looks (status as of May 2026).

The Real-Time canvas runs a diffusion pipeline fast enough that the image redraws while you sketch or type. You scribble a few shapes for a fishing village, the canvas paints a fishing village over them in seconds. Adjust a stroke and the painting follows. It is the quickest path to brainstorming composition and atmosphere without paying for one slow render at a time. For a finished still you usually finalize in Krea Image with Krea 1 or FLUX rather than shipping the Real-Time output directly.

Krea 1 is Krea AI's in-house image model, launched in early 2025. It is tuned for an editorial, slightly tactile aesthetic that the team felt the public APIs were not hitting consistently. Use Krea 1 for portraits, product shots, and editorial illustration where mood matters. For text accuracy, switch to Ideogram inside Krea. For painterly or fantastical scenes, FLUX often still feels more expressive. For vector design, route to Recraft. The point of Krea is that all of these live in one canvas.

Lead with the subject, then the lighting, then the lens or medium. A line like portrait of a chef in a wood-paneled kitchen, side light from a window, 50mm, slight grain reads cleaner than a long descriptive paragraph. Avoid stacking adjectives. Pick one mood word (cinematic, editorial, painterly) and let it carry the look. If a result feels generic, change the medium phrase first, not the subject. Switching 50mm to Polaroid or watercolor often does more work than rewriting the subject.

Krea routes you to several video providers. Veo handles realistic motion and physics well, so reach for it when the clip needs to look like footage. Kling is built for stylized motion plus expressive characters. Pika excels at short effects, transitions, and Pikaffects. Hailuo and Luma fit punchy short cinematic clips. Wan suits stylized loops. The trick is to start from a strong still in Krea Image, then send it into the video model whose look matches the job, rather than re-rolling text-to-video repeatedly.

Krea Train fine-tunes a small adapter (a LoRA) on a base model so it learns a personal style or a recurring character. You upload a set of consistent images, pick a base like FLUX or SDXL, and Krea trains in the background. Six to ten consistent images is enough for a workable style, and twelve to twenty improves character consistency. Pick a short, unusual trigger word like aiko_char and lead with it in every later prompt. Treat the trigger as a key that unlocks the trained look.

Stages lets you set the geometry of a scene before the image model fills the look. You position a product or a figure in a 3D-feeling space, set the camera angle and rough depth, then prompt the surface and lighting. It is closer to staging a virtual photo shoot than writing a paragraph. Stages earn the click when an exact angle matters: a product hero at 45 degrees, a character at low angle on a rooftop, a wide lens on a runner mid-stride. Without an angle requirement, plain Krea Image is faster.

Paid Krea subscribers get commercial usage rights to their outputs, while the free tier ships with usage restrictions. The exact terms are set by Krea AI and by the underlying model providers. Practically that means if you are shipping client work or selling a product, you should be on a paid plan and you should check krea.ai/terms for the current version. The underlying models matter too: Veo, Kling, Recraft and others impose their own licenses, which Krea surfaces inside the relevant generation flows.

Midjourney still leads on painterly mood, especially in v7, and its Discord-and-web-hybrid social loop is unique. Krea's edge is the multi-model canvas: one place to switch between Krea 1, FLUX, Ideogram, and Recraft based on the job, plus real-time iteration, training, and a video module on top. If you only ever generate stylized stills, Midjourney is a strong single-tool pick. If your work crosses image, video, training, and staging, Krea is built to keep all of that in one workflow.

Real-Time outputs are tuned for speed, not final quality. They are perfect for choosing composition, lighting direction, and rough atmosphere, but you should treat the resulting image as a sketch, not a deliverable. Once you like a result, lock the seed if Krea exposes it, then regenerate the same prompt and composition in Krea Image at higher resolution with Krea 1 or FLUX. You will keep the look and pick up the detail, contrast, and texture the real-time pipeline trades away.

Yes. Krea Image, Stages, and Video all accept uploaded reference images, and the canvas treats them like any other layer. Common moves: drop a product photo into Krea Image and re-light it, drop a sketch into Real-Time to paint it up, send a finished still into Krea Video for an animated clip, or use a screenshot as a depth reference inside Stages. Krea is built around image-to-image and image-to-video flows, which usually outperform pure text generation when you start with any reference at hand.

Open Real-Time first. Scribble three shapes for the rough composition, type ten words, watch the painting take shape. Move on to Krea Image with the same prompt to finalize, and try Krea 1 first, then swap to FLUX if you want a more expressive feel. If a result needs more text accuracy, switch to Ideogram inside Krea. If you ship a lot of variations, train a personal style in Krea Train once and call its trigger word from now on. That four-step loop covers the majority of paid creative work.