AI image and video models are remarkably capable, but they only know what you tell them. A prompt like "a city at night" leaves every important decision to chance. A structured prompt turns those decisions into choices - and makes your results repeatable.
The structure of a good image prompt
Build prompts from the same building blocks every time, in roughly this order:
- Subject - who or what, with specific details: "an elderly fisherman with a white beard and a yellow raincoat".
- Action - what the subject is doing: "mending a net".
- Setting - where and when: "on a wooden pier at dawn, light fog over the harbour".
- Composition - framing and angle: "close-up portrait, eye level, shallow depth of field".
- Lighting - "soft golden backlight", "harsh midday sun", "neon reflections".
- Style - "documentary photography", "watercolour illustration", "3D render", "film still".
- Mood and colour - "calm, nostalgic, muted blue and orange palette".
Weak: "a fisherman"
Strong: "close-up portrait of an elderly fisherman with a white beard and a yellow raincoat mending a net on a wooden pier at dawn, light fog over the harbour, soft golden backlight, documentary photography, shallow depth of field, calm and nostalgic mood, muted blue and orange palette"
What changes for video prompts
A video prompt needs everything an image prompt has, plus time. Add three things:
- Motion of the subject - "slowly turns towards the camera", "waves crash against the rocks".
- Camera movement - "slow dolly in", "aerial drone shot moving forward", "handheld tracking shot", "static locked-off shot".
- Pacing - "slow motion", "gentle", "fast-paced", matched to the clip length.
Example: "Aerial drone shot slowly moving forward over a misty pine forest at sunrise, sun rays breaking through the fog, birds flying across the frame, cinematic, calm, 16:9"
Tip: Describe one main action per clip. Short clips with one clear movement look far more convincing than a clip that tries to tell a whole story.
Start from an image for more control
When the exact look matters - a product, a character, a brand style - generate or upload a still image first and use it as the starting frame for the video. The model then animates your composition instead of inventing a new one. In Pynokio, Image and Video work with leading generative models, so you can pick the right model for each step.
Choose the right format
| Aspect ratio | Use it for |
|---|---|
| 16:9 | YouTube, websites, presentations |
| 9:16 | Shorts, Reels, TikTok, Stories |
| 1:1 | Feeds, profile visuals, thumbnails for marketplaces |
| 4:3 | Classic photo framing, slides |
| 21:9 | Cinematic banners and wide headers |
Decide the format before you generate. Cropping a horizontal image into a vertical one usually cuts away the subject.
Keep a series consistent
For a video series, a brand or a faceless YouTube channel, consistency matters more than any single image:
- Create a style block - a fixed piece of text for lighting, palette and style that you paste at the end of every prompt.
- Describe recurring characters identically every time: the same age, hair, clothing and distinctive details.
- Use the same model and format for the whole series.
- Use Templates on the Pynokio home page to apply a ready-made style to your own photo with one click.
Use negative prompts sparingly
Where a model supports it, a negative prompt lists what to avoid: "text, watermark, extra fingers, blurry". Keep it short and specific. Long negative prompts rarely help and can make results less predictable.
Iterate deliberately
- Start with the structured prompt.
- Generate several variations and pick the closest one.
- Change one element at a time - the lighting, then the angle, then the style - so you learn what each word does.
- Save prompts that work. A personal prompt library is the fastest route to consistent results.
Prompt examples to adapt
- Product shot: "matte black stainless steel water bottle on a wet slate surface, water droplets, dramatic side lighting, dark background, commercial product photography, sharp focus"
- Illustration: "cozy reading nook by a rainy window, cat sleeping on a pile of books, warm lamp light, gouache illustration, soft textures, autumn colours"
- Video b-roll: "slow dolly in on a barista pouring latte art in a sunlit cafe, steam rising, shallow depth of field, warm tones, cinematic"
- Thumbnail background: "futuristic city skyline at dusk, glowing purple and teal neon, empty space on the left side, high contrast, digital art"
Use generated media responsibly
Do not create realistic images or videos of real people doing things they did not do, and disclose realistic synthetic content when publishing it. Generated files in Pynokio come with a machine-readable provenance record - see our AI Transparency notice.
FAQ
How long should an AI image prompt be?
Long enough to cover subject, setting, lighting and style - usually one to three sentences. Extra words that do not describe something visible rarely help.
Why do my AI videos look unnatural?
Usually because the prompt asks for too much motion or too many actions. Describe one clear movement and one camera move per clip, or start from a still image.
How do I keep the same character across images?
Describe the character with exactly the same details every time, keep a fixed style block, use the same model, and use an image of the character as a reference where possible.
Which aspect ratio should I use for Shorts and Reels?
9:16 vertical. Choose it before generating so the composition is framed for vertical viewing.
Put these prompts to work. Generate images and video with leading models from one account.
