Target Guide: image to video prompt

How to Write an Image-to-Video Prompt: The One-Action Rule (Stop AI Morphs)

When animating a still image in AI video models (like Botock Video Generator, Kling 1.5, or Sora), the most common mistake is asking for multiple consecutive actions. Here is how the One-Action Rule guarantees photorealistic output.

By Botock AI Filmmaking LabOctober 20, 202611 min readTested on Botock AI Video Generator

The Golden Principle of Image-to-Video

An image-to-video model does not know what happened before the picture, nor what should logically happen five minutes later. It only has 4 to 10 seconds of pixel diffusion budget. If you ask for three events, it blends them together into rubbery morphs. One clip must equal exactly ONE physical change.

Action Density vs. Video Quality Outcome

Prompt TypeWhat You WroteModel OutputFailure Risk
Multi-Action (Fatal)She stands up, walks to the window, and opens itSubject floats, arms melt into window sill, torso stretches95% Glitch Rate
Compound ActionHe nods thoughtfully while smiling and raising a glassMouth twitches violently, extra fingers appear on glass70% Glitch Rate
One-Action Rule (Perfect)The woman turns her head smoothly from frame-left to frame-right. Nothing else moves.Crisp photorealistic head rotation, hair physics stable, face locked0% Glitch Rate

Avoid vs. Correct: Production Prompt Comparison

AVOID (Glitches & Morphing)
"She stands up from the wooden chair, walks quickly over to the balcony door, opens the lock with her key, and looks outside at the night sky smiling."
Why it fails: Four actions crammed into 5 seconds. The AI will drop the character through the floor or morph the chair into the balcony door.
CORRECT (Production Ready)
BASE IMAGE: Maintain exact character identity, lighting, and wardrobe from source image.
ACTION: Person A turns head slowly from looking down at the desk to looking up toward frame-right over 3 seconds and stops. Body, shoulders, and chair remain completely motionless.
CAMERA: Tripod locked-off, zero tilt, zero pan.
Why it works: Single atomic physical displacement. Physics engine preserves all facial features and lighting.

Production Step-by-Step Implementation

Step 1: Freeze 90% of the Scene

Always state what stays completely still: 'Wardrobe, chair, and background remain totally static.'

Step 2: Define the Moving Anatomical Part

Name only one limb or joint: 'Only the right arm moves' or 'Only the eyes shift direction'.

Step 3: Specify the End Frame Stop

Models love to loop actions forever. Always append '...and comes to a complete rest at second 4.'

Step 4: Stitch in Timeline

If you need the character to walk to the window, make that Clip #2 using the final frame of Clip #1.

Interactive Studio

Test This Prompt in Botock Video Generator

Ready to generate your clip? Launch the Botock AI Video Generator right now with zero software installation and test these exact physical constraints.

Launch Video Generator