▶ Short-Form Video · Emotional Micro-Story

THE RED BIKE a tiny story in 90 seconds

A grandfather, a granddaughter, and a red bike — a warm setup, a quiet turn, and an emotional payoff. The format lives or dies on one thing: the same people in every frame. This page shows both how it lands when you lock the cast, and how it falls apart when you don't.

format
9:16 vertical short
scenes
5 intimate beats
image model
flux-dev
consistency
anchored cast-lock
tested verdict
~85–90%
An intimate story dies if "he" looks like three different people. That is the default failure mode of AI micro-stories, not an edge case. The five scenes below are the ANCHORED set — name the cast, pass an explicit character_anchor URL, and restate the wardrobe in every prompt. Scroll down to the consistency lesson to see what the default auto-anchor did instead.

Finished reel — the stitched short

All 5 anchored scenes · 50s · animated on seedance-i2v, warm inworld-tts narration over a looped, ducked piano score — full-length audio, no silent tail.

The 5 beats — the anchored (good) set, rendered on flux-dev

Each beat is a native 9:16 still from flux-dev, with the same grandfather and granddaughter held across every frame via the cast-lock recipe. Slow, lingering, warm — setup → turn → payoff.

Grandfather gives the red bike
01 · SETUP
The gift
Grandpa wheels out a red bike for the little girl in the grey shirt. Her first one. A warm, ordinary morning.
Learning to ride together
02 · SETUP
The lessons
The SAME grandfather runs alongside, one hand on the seat. Summer after summer. Scraped knees and laughter.
She rides on her own
03 · TURN
Letting go
He lets go. She rides on her own for the first time, looking back — and he's smaller in the distance now.
The bike in the garage years later
04 · PAYOFF
Years later
The red bike, rusted, in the garage. The same hands — older now — resting on the handlebars one last time.
She passes the bike to her own child
05 · PAYOFF
The handoff
She wheels the same red bike out for her own child. The grey shirt, the same smile. The story comes around.

Consistency lesson — the default auto-anchor vs the cast-lock recipe

Same prompt, same model, same scenes — the only difference is whether the cast was locked. On the left, the default auto-anchor. On the right, the anchored recipe. This is the single most important decision in the whole format.

Before — default auto-anchor FAILED

Drift scene 1 Drift scene 3 Drift scene 5
The auto-anchor failed to extract the characters from terse prompts and let them drift hard: a 7-year-old aged into a teenager and the grandfather's identity was swapped entirely by the final frame. Three different children, two different grandfathers. The story is broken.

After — cast-lock recipe ~85–90%

Anchored scene 1 Anchored scene 3 Anchored scene 5
Name the cast + pass an explicit character_anchor URL (a real URL, not "auto") + restate the wardrobe every scene. The child's age, the silver hair, and the red bike all held across the reel (~85–90%). Same girl, same grandfather, one story.

Tested verdict — honest, not inflated

~85–90% flux-dev · cast-lock recipe default auto-anchor FAILED

With the cast-lock recipe, ~85–90%. Name the cast, pass an explicit character_anchor URL (not "auto"), and restate the wardrobe in every scene prompt. The anchor holds the face, age, and key props; the prompt holds the clothes. The result is a coherent, intimate story with the same people throughout.

The DEFAULT auto-anchor FAILED — it aged a 7-year-old into a teenager and swapped the grandfather's identity entirely by the final frame. Never ship an emotional narrative on the auto-anchor alone; that's the default failure mode, not an edge case.

For a 95% hero series, train a per-character /lora and apply it to every scene. The anchor recipe is the fast ~85–90% path; the LoRA is the gold standard.

Pipeline trace — the cap chain this reel ran

Step 0 — cast lock (MANDATORY) render ONE clean reference frame → pass as character_anchor: <url> (real URL, NOT "auto") restate wardrobe EVERY scene — "the SAME grandfather in the grey polo", "the SAME girl in the grey shirt"
Wave 1 — scene keyframes 5 × generate_project image=flux-dev, aspect=9:16, named cast + anchor → 5/5 done, identity held
Wave 2 — finishing (per playbook) ltx-q-i2v / seedance-i2v slow gentle motion → chatterbox-tts warm narrator minimax-music "warm ambient piano, slow" → ffmpeg-burn-subtitles ffmpeg-concat + ffmpeg-audio-mix + ffmpeg-export 9:16

Reproduce this — the playbook is open

Built from the emotional-microstory playbook — point it at any small human moment ("the last voicemail she never deleted", "the dad who learned to braid hair") and it writes setup → turn → payoff, then renders one warm intimate scene per beat. Do Step 0 — lock your cast — first, every time.