Prompting
Practical prompt advice for z-image-turbo, flux-2-klein-4b and wan-2.2-i2v (coming soon), with example prompts, and how seed, aspect ratio and resolution affect the result.
This page is practical advice for the two live models. None of it is a guarantee: the same prompt can work well on one seed and badly on another, so plan to try a few. Images are cheap (0.08 credits at 1k), so iterate on the image before you spend credits on video.
Windpaint sends your prompt to the model as written. There’s no prompt rewriting and no negative prompt field. If you want a ready-made style or camera move, use a preset, which wraps your prompt in a published template.
Images with z-image-turbo
z-image-turbo follows plain, descriptive sentences. It does better with a specific description than with a list of loose keywords.
Put the subject first. Start with what the image is of, then add the setting, then lighting, camera and style. The model pays most attention to what comes early, so lead with what matters most.
Be concrete. “A woman” gives the model little to work with. “A woman in her sixties with short grey hair, wearing a yellow raincoat” gives it a lot. Name materials, colors, counts and positions when they matter.
Name the medium. Say “studio product photo”, “35mm film photo”, “watercolor illustration” or “flat vector icon”. Without it, the model picks one.
Describe lighting. Lighting changes an image more than most details do. “Soft window light from the left”, “overcast daylight”, “hard noon sun with short shadows”, “neon light reflecting on wet pavement”.
Say what you want, not what you don’t. There’s no negative prompt, and writing “no people” can make people more likely, because the word is in the prompt. Describe the scene you want instead: “an empty street at dawn”.
Put text in quotes. If you want words in the image, quote them exactly and say where they go: a hand-painted shop sign that reads "OPEN LATE". Short text works better than long text. Check the result; letters can still come out wrong.
Keep it to one scene. A prompt that describes several unrelated things tends to blend them. If you need two ideas, make two images.
Example image prompts
| Use | Prompt | Settings |
|---|---|---|
| Product shot | A matte black ceramic coffee mug on a pale oak table, studio product photo, soft light from the left, plain light grey background, shallow depth of field | 1k, 1:1 |
| Editorial photo | A fisherman mending a green net on a wooden dock at dawn, low fog over the water, muted colors, 35mm film photo, natural light | 1k, 3:2 |
| Illustration | Flat vector illustration of a bicycle courier riding through a city at night, limited palette of navy, orange and cream, clean shapes, no outlines | 1k, 4:5 |
| Text in the image | A vintage enamel sign on a brick wall that reads "WINDPAINT HARDWARE", red letters on a cream background, slight rust at the edges, daylight | 1k, 16:9 |
| Start frame for video | A paper boat floating on a rain-soaked city street at dusk, puddles reflecting warm shop lights, low camera angle, cinematic photo | 1k, 16:9 |
Video with wan-2.2-i2v (coming soon)
Video generation isn’t live yet. This is how to prompt wan-2.2-i2v once it is.
wan-2.2-i2v animates a start frame you supply. The frame already decides what’s in the shot, so the prompt’s job is to describe what moves and how. Restating everything in the frame adds little.
Describe the motion of the subject. Use clear verbs and say how fast: “she slowly turns her head toward the window”, “the dog shakes water off its fur”, “the flag ripples in a steady wind”.
Describe the camera separately. Name one camera move, or say the camera is still. Terms the model tends to follow:
| Camera term | What it does |
|---|---|
static shot, locked-off camera | Camera doesn’t move. Only the subject and scene move. |
slow push in, dolly in | Camera moves toward the subject. |
pull back, dolly out | Camera moves away and reveals more of the scene. |
pan left / pan right | Camera turns horizontally from a fixed point. |
tilt up / tilt down | Camera turns vertically from a fixed point. |
tracking shot | Camera follows a moving subject. |
orbit around the subject | Camera circles the subject. |
handheld | Small, natural shake. |
One camera move per clip works better than a sequence. A 5 second clip doesn’t have room for “pan left, then zoom in, then tilt up”.
Add ambient motion. Small movement in the environment makes a clip feel alive: “steam rising”, “leaves drifting”, “rain falling”, “curtains moving in a breeze”, “light flickering”.
Keep the start frame consistent. The model works best when the motion you describe fits what’s in the frame. Don’t ask for things that aren’t there (“a second person walks in”) or for changes the frame can’t support (“she’s now wearing a red dress”). Large changes tend to make faces and objects drift away from the start frame. If you want a different scene, make a different start frame.
Keep motion modest. Five seconds is short. Gentle, continuous motion usually looks better than fast or complex action, which is more likely to blur or distort.
Match the aspect ratio to the frame. The clip’s size comes from aspect_ratio, not from the start frame. Use the same aspect_ratio you generated the frame with, or the closest of the ten for an uploaded image. See Video.
Example video prompts
| Start frame | Prompt |
|---|---|
| Coffee mug product shot | Static shot. Steam rises slowly from the mug and curls to the right. Soft light stays constant. |
| Fisherman on a dock | Slow push in. The fisherman pulls the net across his lap with both hands. Fog drifts slowly over the water behind him. |
| Paper boat on a wet street | Low tracking shot following the boat as it drifts forward. Rain falls steadily, ripples spread around the boat, reflections shimmer in the puddles. |
| Portrait | Static shot. She blinks, then slowly turns her head toward the window and smiles slightly. Her hair moves a little in a light breeze. |
| Landscape | Slow pan right across the valley. Clouds move across the sky and their shadows pass over the hills. |
Seed, aspect ratio and resolution
A seed fixes the random starting point of a render. The same prompt, model, resolution, aspect ratio, inputs and seed give you the same output. Change any one of those and you get a different result, even with the same seed.
- Changing the aspect ratio changes the composition. The model starts from noise the size of the output, so a different size is a different starting point. A seed that gave you a good image won’t give you the same image at a different ratio. It will usually be a related but different picture.
- Keep the seed fixed while you edit the prompt. Then differences between renders come from your wording, not from chance.
- Change the seed to explore. Run the same prompt with several seeds to see a range of results, then keep the seed you like. Each run is a separate job and is billed separately. See Seeds.
- Choose the ratio for the shot. Pick the aspect ratio for the composition you want instead of describing it in words. Wide scenes read better at
16:9or21:9, portraits at4:5or9:16. Price doesn’t depend on aspect ratio. - Compare models on one seed. The same prompt and seed on
z-image-turboandflux-2-klein-4bgives two different takes on the image; pick the model that suits it. - Video uses a seed too. For
wan-2.2-i2v, the same start frame, prompt, aspect ratio and seed usually reproduce the clip, though video renders can vary slightly between runs. If a clip has an artifact you don’t like, try another seed before you rewrite the prompt.
The API doesn’t report the seed it picked when you leave seed out, so always pass one if you might want to reproduce a result.