Plan B

Sep 6, 2026, 8:49 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?

Turn the uploaded portrait ${portrait} into a polished vertical “Plan B” slideshow video that matches the reference Reel format: one full-screen plan image at a time, changing in sequence. Never create a grid, collage, split screen, contact sheet, or multi-panel image.

PLANNER-ONLY CLASSIFICATION AND PROMPT RESOLUTION
First inspect the portrait yourself and choose exactly one internal creative branch for this fictional transformation: male-presenting or female-presenting. This is a visual creative-routing choice, not a claim about the person’s actual gender identity. Use all directly visible presentation cues together, including face, hair, clothing, grooming, silhouette, and styling. Never choose neutral or ambiguous. If cues conflict, route by the dominant overall styling; a subject visibly presented in a feminine dress or gown must use the female-presenting branch, while a clearly male-presenting subject must use the male-presenting branch. Also decide whether the uploader appears older (roughly 55 or above) or younger/middle-aged. These decisions belong to the planner, not the image generator. Resolve them before calling any image model.

Lock that single branch for the entire run. Before issuing calls, cross-check that PLAN A’s partner, PLAN B’s entourage, and PLAN C’s companion all match the same selected branch. For the female-presenting branch, every image request must explicitly call the uploader “the uploaded woman”; for the male-presenting branch, explicitly call the uploader “the uploaded man.” Do not mix or soften the selected branch later in the run.

Never pass a conditional such as “if the uploader appears older,” “depending on gender,” or “choose an appropriate companion” to an image model. Every image-model request must describe one already-selected scenario. In particular, every PLAN C request must explicitly state the uploader’s selected presentation, the companion’s gender presentation, a concrete visible age target, and the intended age gap. Never ask the image model to classify the uploader or choose which branch to render. Do not mention these classifications in the final output.

Create exactly FOUR separate full-frame images with GPT Image 2 Edit: one image for PLAN A, one for PLAN B, one for PLAN C, and one for PLAN D. Call GPT Image 2 Edit once per plan and use the uploaded portrait as the sole identity reference for every call. For every image set Image Size to portrait_16_9 and Quality to medium.

Make each image a believable, scenario-specific vertical phone photo or cinematic selfie—not the original person pasted unchanged into a new background. Preserve identity through stable facial anatomy and distinctive features: face shape, eyes, nose, mouth proportions, apparent age, natural skin tone, and hair identity. Do NOT preserve or copy the source expression, gaze, facial muscle state, makeup treatment, facial highlights and shadows, white balance, color grade, or apparent light source. Do not paste, composite, or reuse the source face pixels as a cutout.

Before each image call, choose and state in that image prompt a scenario-specific emotional beat, gaze direction, facial expression, facial lighting, and color treatment. Re-light and re-grade the uploader’s face as if they were physically photographed inside that environment: match ambient color spill, key-light direction and softness, shadow density, highlights, contrast, exposure, and depth of field. Preserve natural skin tone underneath the environment’s light, but allow believable warm, cool, neon, golden-hour, candlelight, stage-light, or action-scene color casts. The four images must have visibly different expressions and facial lighting from one another; if any two reuse the same neutral half-smile, gaze, or facial color treatment, regenerate the weaker image.

Use this emotional progression unless a specific scene demands an even better fit:
- PLAN A: warm, relaxed, affectionate joy—genuine smile or candid laugh, engaged eyes, soft welcoming light.
- PLAN B: bold social confidence and exhilaration—playful or celebratory expression, active eye contact with the group or camera, glamorous nightlife or flash lighting.
- PLAN C: poised, intimate sophistication—subtle knowing smile or affectionate gaze toward the companion, elegant cinematic light appropriate to the luxury setting.
- PLAN D: the strongest reaction for the chosen wildcard—determination, awe, adrenaline, triumph, serenity, or delighted absurdity—with dramatic scene-specific light and color. Never default to the same selfie smile used in another plan.

Also re-pose the uploader, adjust body orientation and framing, choose scenario-appropriate clothing, makeup/grooming, and match the camera angle and crop to the environment. Added people must share the same light direction, perspective, scale, depth of field, and photographic treatment. The uploader should remain prominent and immediately recognizable, but does not need to be centered or keep the original clothing, pose, expression, lighting, or color palette. Each image must fill the entire 9:16 frame and show only one plan. Add exactly one label near the top center in bold white text with a thin black outline: PLAN A, PLAN B, PLAN C, or PLAN D. Add no other text, logos, brands, watermarks, borders, UI, or captions.

PLAN C AGE-GAP RULE
PLAN C must show an unmistakable adult age gap; a companion who merely looks a few years older or younger is invalid. Do not change the uploader’s apparent age to manufacture the gap.
- If the uploader appears younger/middle-aged (under roughly 55), choose a specific companion age in the 70–85 range that is visibly at least about 25 years older than the uploader.
- If the uploader appears older (roughly 55 or above), choose a specific companion age in the 25–40 range that is visibly at least about 20 years younger than the uploader. The companion must always be clearly at least 25.
In the PLAN C image prompt, write the selected companion explicitly—for example, “one elegant woman who visibly appears approximately 78 years old” or “one glamorous man who visibly appears approximately 32 years old.” Never include both alternatives. After generation, inspect the result. If the age gap is not immediately obvious at a glance, regenerate PLAN C with stronger visible age cues while keeping the uploader’s identity and apparent age unchanged.

USE THESE SCENARIOS
MASCULINE-PRESENTING:
- PLAN A — polished, comfortable family life with an attractive adult female partner and children.
- PLAN B — glamorous bachelor fantasy in a nightlife, champagne, sports car, yacht, or luxury party setting. The uploaded man must be the ONLY adult man in the image, surrounded by at least three distinct, clearly visible glamorous adult women. A scene with fewer than three women, any other adult man, or the uploader alone is invalid and must be regenerated.
- PLAN C — provocative luxury age-gap romance in an estate, limousine, yacht, or formal-event setting. For a younger/middle-aged uploader, use one sophisticated wealthy woman with the concrete 70–85 age target selected above. For an older uploader, use one glamorous younger adult woman with the concrete 25–40 age target selected above. The planner must resolve this before the image call.
- PLAN D — use the visual-persona seed procedure below to choose exactly one unmistakably extreme cinematic wildcard.

FEMININE-PRESENTING:
- PLAN A — polished, comfortable family life with an attractive adult male partner and children.
- PLAN B — glamorous male-entourage fantasy in a nightlife, champagne, luxury car, yacht, or party setting. The uploaded woman must be the ONLY adult woman in the image, surrounded by at least three distinct, clearly visible stylish adult men. A scene with fewer than three men, any other adult woman, or the uploader alone is invalid and must be regenerated.
- PLAN C — provocative luxury age-gap romance in an estate, limousine, yacht, or formal-event setting. For a younger/middle-aged uploader, use one sophisticated wealthy man with the concrete 70–85 age target selected above. For an older uploader, use one glamorous younger adult man with the concrete 25–40 age target selected above. The planner must resolve this before the image call.
- PLAN D — use the visual-persona seed procedure below to choose exactly one unmistakably extreme cinematic wildcard.

PLAN D VISUAL-PERSONA SEED
Plan D must be selected from the portrait rather than from the planner’s free-form preference. Before choosing it, create a silent visual-persona fingerprint using only non-sensitive visible creative cues: clothing formality and style, dominant colors, pose, gaze, expression, existing setting, lighting, and overall visual energy. Do not infer ethnicity, nationality, religion, sexuality, disability, occupation, wealth, or personality, and do not use gender alone to select the wildcard.

Choose the strongest matching visual archetype below. If several match, break the tie using the most visually dominant cue in the portrait, such as its strongest garment color, pattern, pose, or setting. Then turn that archetype into one concrete, extreme cinematic scenario:
- athletic or kinetic cues → champion athlete, extreme racer, or high-stakes action competitor at the winning moment;
- artistic, edgy, dark, or rebellious styling → tattooed stadium rock star or superstar DJ before a huge crowd;
- serene, minimalist, meditative, or nature-centered cues → monk-like contemplative figure at a spectacular mountain temple;
- warm, playful, affectionate, or animal-associated cues → joyful owner surrounded by an enormous, visually overwhelming dog or cat sanctuary;
- bold, theatrical, heroic, or power-pose cues → superhero in action or mythic champion in an epic setting;
- rugged, practical, or outdoor cues → polar expedition leader, dramatic mountain rescuer, or cinematic wilderness adventurer;
- colorful, quirky, whimsical, or highly expressive cues → surreal ringmaster, larger-than-life magician, or another playful spectacle;
- futuristic, metallic, technical, space-associated, or exploratory cues → astronaut in space.

Astronaut is valid only when at least two visible cues specifically support the futuristic/technical/exploratory archetype. Otherwise choose another category. The final image request must state exactly one already-chosen wildcard scenario, never a list of options or an instruction for the image model to choose. The scenario must strongly contrast with Plans A–C and be instantly readable through extraordinary costume, setting, scale, crowd, transformation, or number of animals. Reject ordinary jobs, workshops, hobbies, homes, vacations, and everyday parties. Never combine multiple wildcard concepts or repeat an earlier scenario.

Keep all added people clearly adult except children in PLAN A. Use candid phone-photo realism, believable anatomy, and a playful aspirational tone. Never sexualize or degrade anyone.

After all four images succeed, call Images to Video exactly once. Pass the four generated images in this exact ordered list: [{url: PLAN A image URL, frames: 48}, {url: PLAN B image URL, frames: 48}, {url: PLAN C image URL, frames: 48}, {url: PLAN D image URL, frames: 48}]. Set FPS to 24. Do not reorder, omit, duplicate, or reuse any plan image. The final video must be exactly 8 seconds, silent, vertical, and show each full-screen image for exactly 2 seconds in the order PLAN A, then PLAN B, then PLAN C, then PLAN D. Use hard cuts only; do not animate, morph, pan, zoom, crossfade, overlap, or show multiple plans together. Do not call Compose Media or any generative video model. Do not ask the runner questions. Return ONLY the final slideshow video as the output. Do not include, attach, list, or expose the four intermediate PLAN A, PLAN B, PLAN C, or PLAN D images in the final response or final outputs.

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 Luna

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

Declare the fields every run must return. You define the shape; the agent fills in the values.

Field

Type

What goes here

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?

Keep all added people clearly adult except children in PLAN A. Use candid phone-photo realism, believable anatomy, and a playful aspirational tone. Never sexualize or degrade anyone.

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 Luna

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

Declare the fields every run must return. You define the shape; the agent fills in the values.

Field

Type

What goes here