Neighborhood Champions

Sep 24, 2026, 6:24 PMRead-onlyPublic

Public page

Configure

Inputs

What the person running this agent provides

Inputs are collected before the run starts. Reference them from your instructions as ${variable_name}.

Your portrait

Variable name

Type

Required

Goal & instructions

What should it achieve, and how should it behave?

Create a 15-second vertical, photorealistic mini-documentary called “Neighborhood Champions.” The runner supplies one portrait: ${portrait}. Make the pictured person the recognizable star of all three scenes. The joke is that an ordinary neighborhood day is filmed with the visual grandeur of a football legend profile. The person does not become, resemble, or interact with a real celebrity.

Creative contract:
- Exactly three scenes, about five seconds each, in one coherent lived-in neighborhood. Scene 1: the person takes a quiet morning tea or coffee on a balcony or outside a small café, carrying the calm confidence of a star before a match. Scene 2: they put real care into a hands-on local task such as fixing a bicycle or motorbike, serving at a market stall, or helping at a shop; choose the task that naturally suits the portrait and neighborhood.
- Scene 3 is exactly one tailored wildcard role. Examples for inspiration, not a whitelist: an absurdly celebrated corner-store cashier, a rooftop pigeon-racing champion, a charismatic street chess monarch, an alleyway tailor receiving a hero's welcome, or an impromptu neighborhood referee. Invent a sharper role if it fits this particular person better. Select one coherent concept, not a mashup. It must contrast clearly with scenes 1 and 2, and finish with a playful, memorable visual reveal. Avoid a generic routine errand as the final beat.
- Keep the person’s apparent age, presentation, facial anatomy, skin tone, hair, and distinctive features recognizable. Match wardrobe and activity to their apparent age and presentation. Do not infer or state real identity, nationality, occupation, socioeconomic status, or gender from the portrait. The setting may be an original warm, lively urban neighborhood; do not assume the person is Egyptian or copy an actual location from the source.
- Every scene must show a different action and camera angle. Use natural handheld/tracking movement, foreground depth, active hands and body motion, warm daylight, neighborhood textures, and believable contact with objects. Film like a polished sports documentary with understated humor, not a static portrait slideshow. No football match, football jersey, trophy, celebrity likeness, brand logos, watermarks, captions, or on-screen text. No violence or humiliation.

Execution contract:
1. Use GPT Image 2 Edit exactly three times, each with the original portrait as the input image, to make one realistic 9:16 scene reference image per scene. Describe the setting, wardrobe, action, and camera composition explicitly. Keep the same person and coherent neighborhood look. Generate images in one parallel batch when possible.
2. Use Kling Video V3 Pro Image to Video exactly three times, one per scene reference, at 9:16 and 5 seconds. Each clip must animate the person actually doing the action, with body/hand movement, physical camera travel, perspective change, and environmental motion. Avoid digital push-ins, frozen poses, texture warping, foot sliding, and talking-photo movement. Dispatch the three independent video calls together when possible.
3. Use Stitch Videos once in scene order with hard or motion-matched cuts, no crossfades or titles. Return only the final stitched video and a one-sentence note naming the wildcard role. Do not return the intermediate scene images or clips as public outputs. Stop after one complete attempt; do not silently produce extra scene variants.

Type $ to insert an input variable. References are replaced with the runner's values at runtime.

Planner model

The LLM that plans and coordinates each run

GPT-5.6 Sol

Allowed tools

The models & abilities it may reach for

Guardrails

Set limits on what your agent can do

Budget cap

per run · pauses to ask if exceeded

Max steps

tool calls per run · leave empty for no step limit

Banned actions

Output

The shape the agent should return