Image to Video Prompts: A Practical Guide for 2026

Short answer: A good image-to-video prompt has four blocks: use the image as the reference, name what must stay the same, describe one motion and one camera move, and list the failure modes you do not want. Keep the clip short, test the motion at low resolution, and check the face, hands and product edges at full size.

Why image-to-video prompts are different

A text-to-video prompt describes a whole scene. An image-to-video prompt describes a change to an existing scene. That is a different job. The model already has the subject, the composition, the lighting and the style. Your prompt should protect those and add motion, not re-describe everything.

The most common mistake is writing a text-to-video prompt and attaching an image. The model then receives two competing descriptions: the image says one thing, the prompt says another. It may split the difference, change the subject, or ignore the image. Start with the image as the authority, then add the motion.

The four-block structure

Block 1: Reference

Say exactly what the image is for. “Use the provided image as the exact visual reference and starting composition.” This tells the model that the image is not a mood board; it is the first frame.

Block 2: Preserve

Name the identity, costume, product shape, label, pose and framing. For a person: “Keep the subject’s recognizable face, costume, hairstyle and body proportions consistent.” For a product: “Keep the bottle shape, label position, logo and material unchanged.” For a scene: “Keep the architecture, window positions and perspective unchanged.”

Block 3: Motion

Describe one subject motion and one camera move. “The subject slowly turns their head toward the camera. A gentle breeze moves loose hair. The camera performs a slow push-in.” Or: “Fog drifts from left to right. The lantern flickers. The camera holds steady.” One motion, one camera move.

Block 4: Guardrails

List what must not happen. “No identity change, no new characters, no scene cuts, no text, no watermarks.” For a product: “No label changes, no extra objects, no geometry warps.” Guardrails are not magic, but they reduce the most common failures.

Eight motion types that work in five seconds

1. Head turn. A slow turn toward the camera. Good for portraits; keep the shoulders still.
2. Hair movement. A gentle breeze. Good for costume and beauty shots.
3. Fog drift. Background motion without subject motion. The safest atmospheric effect.
4. Light flicker. A candle or lantern changes the highlights. Good when the subject must stay still.
5. Camera push-in. A slow move toward the subject. Adds drama without changing anatomy.
6. Orbit. A slow move around a product. Good for reflective objects if the label stays visible.
7. Turntable. A product rotates on a clean floor. Good for shoes and packaging.
8. Macro glide. A close move across a texture. Good for skincare, fabric and food.

If you need a walk, a dance, a fight or a dialogue, you are asking for a different workflow. Those actions need more time, more control and often several shots.

What usually goes wrong

Identity drift. The face changes because the motion is too large, the source is low-detail, or the prompt does not protect the identity. Reduce the motion and strengthen the preserve block.

Anatomy warps. Hands, fingers and limbs are the first to break. Keep hands out of the frame or keep them still. If the source has a hand near the face, say “keep the hand placement unchanged.”

Label warps. Product text is hard. Use a slow move, a clean source and an explicit “do not change the label” line. Check the label at full size.

Background melts. Busy backgrounds are harder than simple ones. If the background is complex, ask for a slow camera move and keep the subject still.

Scene cuts. The model may invent a cut when the prompt implies a sequence. Say “one continuous shot, no scene cuts.”

A portrait prompt you can adapt

Use the provided image as the exact visual reference and starting composition. Keep the subject’s recognizable face, costume, hairstyle and body proportions consistent. The subject slowly turns their head toward the camera with a calm expression. A gentle breeze moves loose hair strands. Soft fog drifts through the background. The camera performs a slow, smooth push-in. Maintain realistic motion and stable anatomy. Do not change the identity, replace the costume, add characters, cut the scene, or add text or watermarks.

This is the prompt used on the Halloween video generator page. It works because it gives the model one motion and one camera move, and it repeats the identity protection.

A product prompt you can adapt

Use the uploaded product photo as the exact reference. Keep the product shape, label position, logo, colour and material unchanged. Create a five-second slow orbit shot. A thin beam of warm light travels across the surface. A subtle mist drifts in the dark background. Realistic material texture, stable geometry, controlled highlights. Do not change the text or logo, add objects, cut the scene, or add watermarks.

The structure is the same. The preserve block is about the product, the motion block is about the camera and the light, and the guardrails protect the label.

How to test a prompt without wasting credits

Generate the shortest duration at the lowest resolution that still lets you judge the motion. Watch the first second, the middle and the last frame. If the subject holds, render a higher-quality version. If the subject drifts, change one instruction.

Do not generate five versions of the same prompt. That is not a test; it is a lottery. Change the motion, the preserve line or the source image, and note which change fixed the problem. A small prompt notebook is more valuable than a larger credit pack.

Cost and expectations

A four-second 720p Fast clip starts at 40 credits. The Studio shows the cost before submission. The $4.99 one-time pack includes 50 credits, enough for one test with a little left over. Longer clips, higher resolution and extra inputs cost more.

Results are not guaranteed to match a sample. The model can change the face, the product or the background even with a good prompt. Treat the first clip as a draft. If two attempts fail, simplify the motion or go back to the still.

Frequently asked questions

What is the best image-to-video prompt?

A prompt that says “use this image as the first frame,” names what must stay the same, describes one motion and one camera move, and lists the failure modes to avoid. Specificity beats length.

How long should an image-to-video clip be?

Five seconds is enough for one motion. Longer clips cost more and increase the chance of drift. Generate a short draft first, then extend only if the subject holds.

Why does my face change in the video?

Too much motion, a low-detail source or a weak preserve line. Reduce the action, use a sharper photo, and repeat the identity instruction.

Can I use a text prompt without an image?

Yes. Text-to-video is available. It gives you more freedom but less control over the first frame. If you need a specific person or product, start from an image.

How much does an image-to-video clip cost?

A four-second 720p Fast clip starts at 40 credits. The exact cost depends on duration, resolution and settings, and the Studio shows it before submission.