Pika Prompting Guide: Text, Image-to-Video and Pikaframes

Pika turns a sentence, a still image, or a series of images into a short clip. The prompting rules are the same across all three: one subject, one clear motion, a camera instruction and a look, in that order. Pikaframes adds keyframes, so you decide the start and end (and any stops between) and let Pika animate the journey.

How Pika reads a prompt

Video models respond to motion verbs more than to adjectives. "A woman in a yellow raincoat turns to face the camera as rain starts to fall" gives Pika a subject, an action, a trigger and a mood. "Beautiful cinematic moody rainy portrait, 8k, masterpiece" gives it almost nothing to animate.

A reliable order is:

1. Subject with one or two identifying details.
2. Action: a single verb that can play out in a few seconds.
3. Camera: static, slow push in, orbit, handheld, drone pull back.
4. Setting and light: where, what time of day, what the light does.
5. Style: film stock, animation style, colour palette. Keep this to a phrase.

Put anything you do not want in the negative prompt field rather than writing "no text" in the main prompt; video models tend to read "text" as a request. Use the seed to hold a result steady while you change one thing at a time, and the prompt-adherence or strength control to decide how much Pika should follow your words versus its own idea of a good shot.

Text-to-video versus image-to-video

Text-to-video is the right choice when you are exploring and do not yet know what the frame should look like. Image-to-video is the right choice once you do. An uploaded image fixes composition, colour, character design and lighting; the prompt then only needs to describe motion and camera.

For image-to-video, strip the prompt down. The model already knows what the woman and the raincoat look like, so "she turns towards the camera, slow push in, rain intensifies" is enough. Repeating the whole visual description can cause the clip to drift away from your image as the model tries to re-interpret it. Keep source images sharp and free of text or watermarks, and give yourself a little room at the edges if you plan a camera move.

Pika 2.2 supports clips of up to 10 seconds at 1080p. Test short and low, finish long and high: a 5-second draft tells you whether the motion works, and the full render is for the version you keep.

Pikaframes: keyframe transitions

Pikaframes lets you upload two or more images and have Pika generate the motion between them. The first image is the opening frame, the last is the closing frame, and any in between are waypoints. This is how you get a controlled transformation (a sketch becoming a finished render, a product changing colour, day turning to night on the same street) without hoping a text prompt lands.

• Keep the frames related. Same subject, same rough composition, same camera angle. The more two frames differ, the more the model must invent, and the more likely the middle goes strange.
• Describe the transition, not the frames. Pika can see the images; it needs to know how to travel: "the sketch fills with colour from the centre outwards, camera holds still".
• Use the duration to pace it. A longer clip gives a smoother morph; a short one gives a snap-cut feel.
• Chain carefully. For a longer piece, use the last frame of one clip as the first frame of the next so the join is invisible.

Pikaframes is also a quiet fix for character consistency: generate your two poses as stills first (see consistent characters for the still-image side), then let Pika move between them.

Pikascenes and lip sync in brief

Pikascenes (shown as "ingredients" in some versions) combines several uploaded images into one scene: a character, a prop and a location become a single clip. Give it clean, isolated images of each ingredient and a prompt that says how they relate ("the woman holds the lantern while walking through the archway"). Lip sync takes a clip of a face plus audio or typed text and animates the mouth to match. It works best on a front-facing, well-lit face with little head movement. Both tools follow the same rule as everything else on Pika: clean inputs, one instruction, short tests.

Settings at a glance

Names shift between versions; these are the controls to look for in the current interface.

Try it

An image-to-video prompt for a still of a lighthouse at dusk:

Slow drone pull back from the lighthouse as its lamp begins to rotate, waves roll in below, the sky deepens from orange to violet. Camera smooth and level. Negative: text, watermark, people, blur.

Swap the camera phrase first ("slow push in" or "static shot") to see how much the move changes the feel. Then try "lamp flickers on" instead of "begins to rotate" to learn how literally the model reads your verb.