I Turned My Son's Daydream Into a 15-Second AI Animation

Step-by-step process

Step 1: Turn a photo of your child into a 3D Pixar-style character

Upload a clean head-and-shoulders photo to ChatGPT Images 2.0 inside Higgsfield. Prompt for a Pixar-style 3D character version on a white studio background.

My prompt:

"Convert this photo into a high-quality 3D animated character in Pixar style. Expressive face, soft realistic lighting, clean white studio background. Keep facial features and hair colour faithful to the photo."

Result: Took 3 generations to get one I liked. The first looked too cartoonish. The second got the proportions wrong. The third matched what I wanted.

💡 Parent Insight: Generate at least three versions and pick the strongest. The model gives noticeably different output runs even on the same prompt.

‍

Step 2: Generate the football kit version (front and back)

Take the chosen character, then prompt for two new images: one in an Arsenal kit facing forward, one in the same kit from behind. The back view matters because the storyboard prompt needs both perspectives.

My prompt for the front view:

"Same character, now wearing a full Arsenal home football kit (red and white striped shirt, white shorts, red socks). Standing on a green football pitch, confident pose, white studio lighting."

For the back view, I changed "confident pose" to:

"Standing facing away from camera, looking up at the stadium lights. Show full back of jersey."

Result: Two clean reference images that I'd use in the next step as @image1 (front) and @image2 (back).

Step 3: ChatGPT Images 2.0 vs Nano Banana 2 (which won)

I ran the same character prompts through Nano Banana 2 for comparison. Nano Banana 2 produced lower-quality faces with inconsistent proportions. The character looked off in a way that's hard to describe but obvious when you compare side-by-side.

ChatGPT Images 2.0 nailed the Pixar look on the third try. Worth the extra credits (~7 per generation versus ~1.5 for Nano Banana 2).

If you want the full breakdown of what ChatGPT Images 2.0 does well, see our 6 weekend activities guide which has a downloadable PDF.

Step 4: Build the 10-frame storyboard

This is the most important step. The storyboard tells the video model your camera language: where to push in, where to cut, where to slow down. Skip it and the video drifts.

I used a single ChatGPT Images 2.0 prompt to generate a 2x5 grid of 10 frames. Both reference images were attached.

My exact prompt:

"Create a 2x5 storyboard grid (10 frames) for a 15-second stylized 3D animated short film. Style: high-quality 3D animation, cinematic composition, soft depth of field, expressive character design, realistic lighting, subtle motion blur. Theme: a young boy in @image1 (front) and @image2 (back) dreams he is a world-famous football star scoring a winning goal for Arsenal at the Emirates stadium which is packed, but wakes up in his bedroom and realises it was just a dream. Storyboard frames (left to right, top to bottom): 1. Wide establishing shot of a massive football stadium at night, glowing floodlights, cheering crowd. 2. Medium rear shot of the boy in a professional Arsenal kit standing confidently on the field. 3. Tracking shot of the boy dribbling past defenders, dynamic motion, grass particles flying. 4. Close-up of his focused face, intense expression, stadium lights reflecting in his eyes. 5. Slow-motion action shot of his foot striking the ball toward the goal. 6. Ball flying into the net, net stretching dramatically, crowd exploding in celebration. 7. Hero shot: boy celebrating with arms raised, lens flare, cinematic glow. 8. Hard cut transition frame: same pose but now in bedroom, lighting shifts abruptly. 9. Medium shot of the boy sitting on his bed, confused, soft morning sunlight entering room. 10. Wide shot of cosy bedroom with football posters, ball on floor, boy smiling with quiet determination."

Result: Generated 3 versions. Picked v3. v1 had inconsistent character proportions across frames. v2 nailed the football scenes but the bedroom looked off. v3 was clean throughout.

Step 5: First Seedance attempt blocked by eligibility check

I uploaded v3 to Seedance 2.0 inside Higgsfield, pasted my video prompt (below), and hit "Check eligibility" which returned a fail.

The reason: I'd asked the storyboard to include a poster of "Saka" on the bedroom wall (Bukayo Saka, Arsenal's number 7). Higgsfield's eligibility filter blocks images that reference real public figures by name. The poster text "SAK" was enough to flag it.

Solution: Regenerated the storyboard with generic "football posters" instead of named players. Eligibility check passed on the next attempt.

💡 Parent Insight: The eligibility filter is doing useful work. It's protecting against deepfake-adjacent content that uses real footballers' names and likenesses. Keep posters, jerseys, and signs generic.

‍

Step 6: Seedance Fast vs Seedance 2.0 (the credit decision)

Seedance Fast: 21 credits for a 15-second clip. ~€0.53. Seedance 2.0 Normal: 135 credits for a 15-second clip (currently discounted from 180). ~€3.38.

I tried Fast first to save credits. The output was disappointing on the action shots. The dribble was visual mush. The kick frame had blurry feet. The bedroom transition felt rushed.

Switched to Seedance 2.0 Normal. Action shots clean. The match-cut from the celebration pose into the bedroom landed seamlessly: same arms-raised pose, lighting shift from stadium spotlights to morning sunlight, no jarring jump.

My exact video prompt:

"Use the provided storyboard image as the primary visual reference for composition, framing, and scene progression. Generate a 15-second stylised 3D animated cinematic video that follows the exact sequence and visual storytelling of the storyboard. Style: High-quality 3D animation, cinematic realism, expressive character design, soft depth of field, realistic physics, subtle motion blur, dramatic lighting contrast. Theme: A young boy dreams he is a world-famous football player scoring a winning goal in a packed stadium, but wakes up in his bedroom and realises it was just a dream. Execution: Match all camera angles, framing, and transitions from the storyboard image. Maintain consistent character design between dream and real-world scenes. Smooth cinematic camera motion (tracking shots, slow push-ins, slight rotations). Slow motion during key action moments (dribble, kick, goal). Realistic environmental effects: grass particles, stadium lights, crowd motion. Scene flow (15s timing): 0-5s stadium sequence (wide establishing, dribble, close-up). 5-9s goal moment (kick, ball hits net, crowd erupts). 9-11s celebration hero shot (lens flare, slow-motion). 11-13s hard cut transition (stadium to bedroom, same pose match-cut). 13-15s bedroom reveal (soft morning light, calm mood, subtle smile). Lighting: Dream uses high-contrast stadium lights, cool tones, bright highlights. Reality uses warm, soft natural sunlight, grounded calm atmosphere. Mood: Start epic and high-energy, then shift into calm, intimate, slightly emotional ending."

Result: Render finished in about 90 seconds. The match-cut transition was the standout moment.

Step 7: Show your child

I called Mateo over and pressed play. He watched the full 15 seconds without speaking. Then he ran to his bedroom, came back wearing his Arsenal kit, threw both arms in the air to match the hero shot, and asked me to play it again. He watched it on loop the rest of the evening.

‍