Finding Kaby: A short film with Seedance 2.5 Get my complete...

Finding Kaby: A short film with Seedance 2.5

Get my complete process, with prompts, below 👇

Tools and apps used:

Images & Videos: @higgsfield_ai using GPT Image-2
and Seedance 2.5
Agent: @higgsfield Supercomputer
Brainstorming & prompts: @claudeai Fable 5 Max
Editing: @capcutapp
Additional SFX: Seed Audio 1.0
Music: @epidemicsound

First, I had to upscale the meme images with this prompt:

Create a remastered version of the attached image, preserving the exact character, scene structure, and composition. Keep every element in the same position and at the same scale, with identical poses, expressions, and framing. Improve only the execution quality: sharper detail, cleaner lines, richer and more accurate color, better lighting depth, and higher resolution rendering. Do not add, remove, or rearrange anything in the scene.

I used the newly upscaled image, and using this prompt, I made it a 3D animated character sheet using GPT Image 2:

Stylized 3D character reference card for the young woman from the reference image. Keep the same straight shoulder-length hair, direct wide-eyed stare, and fixed toothy smile. Present as a clean production model sheet with a neutral studio background, readable full-body view, head close-up, and one expression callout. Feature-film-quality 3D animation, polished and consistent.

I repeated the process for all the other characters, then took them to Claude, using Fable 5 Max, where we brainstormed a bit about the story.

It was all done in text-to-video, with references.

The next step was to find a style to be used in all the prompts:

Premium 3D animated feature film look with stylized realism: soft subsurface skin, detailed cloth and hair simulation, cinematic depth of field, filmic warm and teal color grade. The city is dense and alive in every frame: sidewalks carry a steady flow of background pedestrians each absorbed in their own small actions, with layered secondary motion such as passing traffic, pigeons, steam, and shop activity in every shot. Every bound character is a vivid expressive performer with strong readable facial emotion, animated eyebrows, and full body acting, behaving like a real person filmed candidly in continuous natural motion at all times, breathing, blinking, shifting weight, and gesturing; signature meme actions are performed big and clear within and on top of that natural motion, never as held poses; no character ever appears static, posed, flat faced, or photo like. The camera is an invisible documentary observer: no character ever looks into the lens, glances at the camera, or acknowledges the audience in any way; every look, gesture, and line of dialogue is directed at other characters, objects, or events inside the scene. All locomotion is natural and forward: every character faces the direction they are moving, walks and runs facing forward, never backwards or sideways unless the action explicitly states it, and takes stairs facing forward down or up the steps. Audio is strictly diegetic: no music of any kind, no score, no soundtrack; every sound is a diegetic sound effect produced inside the scene.

Here's an example prompt for the first scene, using the Seedance 2.5 syntax and rules:

[Generation Goal]
Generate a 3D animated comedy short that opens under a big yellow title card reading "FINDING KABY". strolls a busy Brooklyn street in a great mood, gets an incoming call from his girlfriend, answers the situation with his trademark palms-out shrug aimed at his own phone and a calm decline tap, and glides into the subway without ever losing his vibe.

[Characters]
corresponds to @ kaby.png. Use only the appearance, short fade haircut, blue tie-dye hoodie with the red circular emblem, and dark pants. Do not use the gray studio background; all views in the image depict the same single person. Performance: easy charisma and a warm confident smile, expressive eyes and lively eyebrows, a light bounce in his walk; his signature move is both palms rolling outward at chest level with a head tilt and a knowing smile, always performed as a living motion and always aimed at a person, object, or situation inside the scene.
corresponds to @ laina.png. Her face appears only as the smiling caller photo on 's phone screen; she is never physically present in this shot. Do not use the reference sheet background, labels, or text panels.

[Global Settings]
Premium 3D animated feature film look with stylized Pixar grade realism: soft subsurface skin, detailed cloth and hair simulation, cinematic depth of field, filmic warm and teal color grade. The city is dense and alive in every frame: sidewalks carry a steady flow of background pedestrians each absorbed in their own small actions, with layered secondary motion such as passing traffic, pigeons, steam, and shop activity in every shot. Every bound character is a vivid expressive performer with strong readable facial emotion, animated eyebrows, and full body acting, behaving like a real person filmed candidly in continuous natural motion at all times, breathing, blinking, shifting weight, and gesturing; signature meme actions are performed big and clear within and on top of that natural motion, never as held poses; no character ever appears static, posed, flat faced, or photo like. The camera is an invisible documentary observer: no character ever looks into the lens, glances at the camera, or acknowledges the audience in any way; every look, gesture, and line of dialogue is directed at other characters, objects, or events inside the scene. All locomotion is natural and forward: every character faces the direction they are moving, walks and runs facing forward, never backwards or sideways unless the action explicitly states it, and takes stairs facing forward down or up the steps. Audio is strictly diegetic: no music of any kind, no score, no soundtrack; every sound is a diegetic sound effect produced inside the scene.
Early morning Brooklyn brownstone street packed with neighborhood life: joggers, dog walkers, kids chasing pigeons, neighbors chatting on stoops, a fruit vendor serving customers, taxis and a delivery cyclist passing, steam from a sidewalk grate, the Manhattan skyline between buildings. Camera grammar: the camera tracks parallel with him from across the street in profile coverage with parallax foreground traffic, one dolly-in from a three-quarter side angle reserved for the shrug, and a locked wide from across the street for the subway descent.

[Stage 1] 0-6 seconds
Primary event: mid-action open, @ kaby.png walks forward along the sidewalk screen left to right, seen in full profile, facing the direction he walks with a light bounce and a warm smile; he nods good morning to two neighbors on a stoop as he passes them and pats a passing dog without breaking his forward stride, while a jogger overtakes him in the same direction and kids scatter pigeons ahead of him; he pulls out his phone mid stride and looks down at it. Overlaid on top of the video for this entire stage, big and centered in the middle of the screen: a title card with the exact text "FINDING KABY" in all capital letters, a simple heavy bold sans serif font, bright yellow, playful, with a slight continuous jiggle, its style matching the 3D animation.
End state: still walking forward, phone raised, his eyes on the screen, street life flowing around him, the title card vanishing exactly on the cut at 6 seconds.

[Stage 2] 6-11 seconds
Primary event: the camera looks over his right shoulder at the device screen while he keeps walking forward: a simple full screen incoming call, the smiling caller photo of @ laina.png filling the top half of the screen, beneath it the contact name reading exactly "Laina" followed by a small red heart emoji, beneath that the label reading exactly "Incoming call", and at the bottom a round red decline button and a round green accept button; the phone buzzes in his hand as his eyebrows rise in amused disbelief.
End state: his eyes stay down on the ringing screen as a grin forms.

[Stage 3] 11-17 seconds
Primary event: from a three-quarter side angle the camera dollies in as @ kaby.png, still walking forward, holds the phone out at arm's length and gives the phone itself his trademark palms-out shrug, both palms rolling outward at chest level with the phone resting in one open palm, head tilted, a big knowing smile breaking into a small laugh to himself; his eyes move only between the phone and the street ahead.
End state: the shrug melts back into his forward stride, grin lingering.

[Stage 4] 17-24 seconds
Primary event: he taps the red decline button with his thumb, pockets the phone, raises his hood in one smooth motion while walking forward, and weaves through the small crowd at the fruit stand ahead of him, steadying a wobbling orange with one finger and answering the vendor's wave with a quick friendly fist bump as he passes, never slowing and never turning around.
End state: clear of the stand and still walking forward, the green railed subway stair entrance just ahead of him with commuters climbing out of it and descending into it.

[Stage 5] 24-30 seconds
Primary event: locked wide from across the street, @ kaby.png turns off the sidewalk to face the subway stair entrance, a green railed stairway with a globe lamp cut into the sidewalk and leading down below street level, and descends the stairs facing forward down the steps, his back to the street, with the same relaxed bounce among other commuters, giving a small palms-up flick to a pigeon perched on the stair railing as he sinks step by step below street level while the street keeps moving above.
End state: the busy stair mouth carries on without him as a train rumbles below.

[Audio]

No music, no dialogue, no narration, no watermarks, no subtitles, and no on-screen graphics other than the single opening title card described in Stage 1.

[Maintain Consistency]
Keep 's facial identity, hairstyle, and wardrobe locked to his reference throughout. He is the only main character; background characters stay incidental, they may exchange brief friendly reactions with him, a nod, a wave, a dog pat, a fist bump, but they never speak and never become main subjects. His relaxed pace and good mood never change at any point. Screen direction and locomotion: he travels screen left to right for the entire piece, always walking forward and facing his direction of travel, and he descends the subway stairs facing forward down the steps with his back to the street. The subway appears only as a street level stair entrance descending below the sidewalk; no train, train cars, tracks, or platform are ever visible in this shot. Phone content appears only on the device screen as a diegetic element, and Laina's face exists only as the caller photo on that screen; she never appears physically in this shot. No character looks at, glances at, or acknowledges the camera at any point. The opening title card reads exactly "FINDING KABY": bright yellow, heavy bold, all capital letters, big and centered in the middle of the screen, with a slight playful jiggle, visible from the very first frame and gone exactly at the cut at 6 seconds, never reappearing. Do not duplicate , and do not introduce additional main characters, overlay graphics, watermarks, or text of any kind outside the phone screen and the single opening title card.

For the second-to-last step, I went into the Higgsfield Supercomputer, and I uploaded all the assets and prompts, and I told the agent to figure out what references correspond to what character and run the prompts through Seedance 2.5.

As an additional step, I asked Supercomputer to generate extra sound effects for each video, using Seed Audio 1.0.

It watched the entire video, and it generated sequences of 30-second audio to be used in the edit.

It worked like a charm!

And in the last step, I put everything into CapCut and edited the film, making it all come together as one.

I added a suitable soundtrack from Epidemic Sound and played a bit with the volumes to duck the music and SFX for the dialogue, and that's it, pretty much!

I hope you enjoyed this, and if you learned something new, let me know! ✌️