Skip to main content
All templates

Prompt template

Toddler Learns to Milk a Cow — Handheld Farm Home Video (Image-to-Video)

A 30-second, ultra-realistic handheld home video of the child from your reference image visiting a countryside barn, failing their first try at milking a cow, then succeeding — with three timed beats, toddler dialogue, and natural farm audio. Shared by @AIwithJessica (GPT Image 2 reference sheet + Seedance 2.5).

Made for GPT Image 2Video templateRevision 3

Demo video

View case

Start from an example

Pick an example to load its exact settings, then make it yours.

  • View case

Make it yours

Only the parts worth changing are shown. The prompt updates as you go.

Reference image

Upload your reference image in ChatGPT or Gemini next to this prompt. Nothing is uploaded here.

@Image 1

Character reference sheet of the child and outfit, attached as Image 1.

Who the child in the reference is, e.g. Korean toddler.

Your prompt

Create a 30-second, 1080p, 16:9, ultra-realistic handheld home-video of the SAME Korean toddler from @Image 1 learning to milk a cow at a traditional Korean countryside farm. CHARACTER LOCK Use @Image 1 as the exact reference. Preserve the toddler's face, dark eyes, small nose, round cheeks, skin tone, short dark bowl-cut hair with soft bangs, and chubby toddler proportions. Same child for all 30 seconds. Do not age up or redesign. WARDROBE LOCK Exact cow onesie from @Image 1: fluffy white sherpa fleece with irregular black spots, hood UP with small tan horns and tan inner ears, brown oval belly patch with stitching, small tail, white plush booties. Real plush fabric with visible fibers, folds, and seams. Child is NOT holding a milk carton. STYLE Live-action, real skin texture, natural hair strands, warm summer afternoon light. Reference is for identity and wardrobe only. No cartoon, CGI, plastic skin, or beauty filter. SETTING Old wooden barn, hay, wooden tools, metal buckets, green fields, several calm cows. CAMERA Always directly in front of the child, like a friend filming while walking backward. Handheld smartphone look: natural shake, imperfect framing, slight autofocus hunting, mild noise. No gimbal, drone, rear shots, or cinematic moves. 0:00–0:10 ARRIVAL The toddler happily toddles toward the camera into the barn, hood bouncing. Spotting the cows, they look into the camera and say in a cute toddler voice: "Today I'm gonna learn how to milk a cow!" A calm cow looks back; the child walks closer. A farmer places a small metal bucket beside the child and gestures toward the cow. End: child beside the cow, farmer nearby, child looking up at the farmer. 0:10–0:20 FIRST ATTEMPT The farmer kneels and demonstrates gently. The toddler tries awkwardly and almost nothing enters the bucket. Looking into the camera, confused: "Huh? Why is it not coming out?" A small giggle. The farmer smiles and shows the movement again. 0:20–0:30 SUCCESS The toddler copies the movement and milk flows into the bucket. Eyes widen; proud smile to camera: "It worked! It's really coming out!" The friend filming reacts, causing a small camera shake. The child pats the calm cow, waves, and says: "Today was so much fun! Bye-bye!" Cut to black. AUDIO Only natural farm sounds: cow moos, bucket clinks, footsteps, fleece rustle, birds, insects. Natural child voice, a warm farmer voice ("That's it, gently now"), and a soft off-camera giggle. No music, narration, or sound effects. CONTINUITY One continuous home-video memory. Same child, onesie, cow, farmer, barn, and lighting throughout. No duplicate characters.

When to use it

Use when you have a character reference sheet of a child in a costume and want a cute, phone-filmed 30-second story clip with a fail-then-success arc and spoken lines.

How to get the best result

  • Attach the character reference sheet as Image 1 (identity + outfit only); the original used a GPT Image 2 turnaround sheet of a toddler in a cow onesie.
  • Keep the timed 0:00–0:10 / 0:10–0:20 / 0:20–0:30 beats so the model paces the story.

Watch-outs

  • Identity and outfit can drift over 30 seconds; a multi-view reference sheet holds them better than a single photo.
  • Some video models ignore 'no music' — re-roll or mute if a soundtrack appears.

Source and attribution

Example cases by
Jessica Collins

Browse more