Image-to-Video: Why You Should Generate the Keyframe First
Last updated: October 12, 2026
Text-to-video asks the model to invent the look of the scene and the motion at the same time. Image-to-video splits that into two steps: you lock in the look with a still image, then you only ask the video model to add movement.
That split is why CineCue produces a keyframe prompt and a separate motion prompt. Each step gets clearer instructions, and you can retry the motion without losing a look you like.
When to use it
- You care about exact composition, styling, or character design.
- You want several clips that share the same look.
- Text-to-video keeps giving you a different scene every time.
- You are making ads where the product must look exactly right.
How to write the prompt
1. Write the keyframe like a photo. Describe subject, framing, lens, lighting, and aspect ratio. Leave out motion words; they do not belong in a still image.
2. Generate several stills and pick the best. Images are fast and cheap compared to video. Choose the strongest frame before spending time on animation.
3. Write the motion prompt separately. Describe only the camera move and how the scene behaves over time. Do not re-describe the subject; the image already shows it.
4. Match the aspect ratio. Generate the keyframe in the same aspect ratio you want for the final video, so the animation does not crop or stretch it.
5. Ask the model to preserve identity. Include a line asking it to keep the subject's features and proportions consistent, and to animate only the camera and environment.
Copy-ready example
Subject: A vintage muscle car parked in a rain-soaked Tokyo alley at night. Generate the keyframe image first, then use it as the start frame for your video with the motion layer as the video prompt. Add the negative block wherever your tool accepts one.
A vintage muscle car parked in a rain-soaked Tokyo alley at night. medium hero shot, subject framed from mid-torso up, heroic and commanding presence, shot on a 35mm anamorphic lens, oval bokeh, horizontal lens flares and gentle barrel character, cyberpunk volumetric neon lighting, saturated magenta and cyan glow, atmospheric haze and wet reflective highlights, 16:9 landscape aspect ratio. Photorealistic, cinematic color grade, high dynamic range, sharp focus.
Camera move: steady forward dolly push-in along the Z-axis, camera translating toward the subject on a locked track, smooth cinematic glide, steady professional gimbal cadence. foreground and background planes separate with dynamic parallax, depth compression building as distance closes. strict frame-to-frame identity lock, subject features, wardrobe and proportions held perfectly consistent. Animate only the camera and environment, not the subject's core features.
Negative: blurry, low quality, distorted, deformed, extra limbs, disfigured, watermark, signature, jpeg artifacts, text errors, bad anatomy, flickering, temporal inconsistency, morphing artifacts, warping, ghosting. Style: cinematic film grain, anamorphic character, high dynamic range, professional color grade.
Common mistakes
- Putting camera motion words in the image prompt.
- Re-describing the subject differently in the motion prompt, which causes drift.
- Mismatched aspect ratios between the still and the video.
- Animating a weak keyframe and hoping motion will fix it.
Try it with your own scene
CineCue builds every prompt layer for you. Open it with this setup, describe your own subject, adjust the options, and copy the result into your AI tool.