Render Epic Odyssey Portrait

Turn a real person into the visual center of an original ancient-expedition epic. Create scale through landscape, weather, light, depth, and narrative direction. Keep the person recognizable; do not treat the input as disposable inspiration.

Use the image-generation/editing tool for the photographic transformation. Use a local layout method for text and before-after presentation so generated text never corrupts the image.

Non-negotiable rules

1. Inspect the source image before editing.
2. Preserve identity: face geometry, skin tone, age range, hairline, hairstyle, body proportions, pose, gaze, hand position, camera angle, and distinguishing accessories.
3. Preserve source facts unless the user explicitly asks to change them. Do not add a spouse, child, crowd member, weapon, crown, injury, tattoo, logo, or religious symbol.
4. Replace or extend the environment only as far as the selected transformation level permits.
5. Build an original classical voyage aesthetic. Never copy a named film's frames, costume, color script, typography, actor likeness, studio branding, or poster composition.
6. Do not use generic fantasy clutter—random runes, sparks, dragons, floating rocks, excessive gold armor, or unexplained magical effects—as a shortcut to scale.
7. Keep source files local and never overwrite them. Do not expose private faces or metadata beyond the user's requested deliverable.

Choose the transformation level

Select the lowest level that can deliver the requested contrast. State it in the transformation card.

• L1 — World expansion (default): keep the person and clothing intact; extend or replace the setting with a monumental voyage landscape and cinematic light.
• L2 — Material translation: preserve garment silhouette, coverage, dominant color, and personal accessories; translate fabric and surface details into an original expedition-era wardrobe. Use only when the user asks for a stronger character transformation.
• L3 — Mythic elevation: allow an original cloak, ceremonial armor, staff, shield, or weathered expedition equipment while preserving identity, pose, and anatomy. Require explicit user intent. Keep props narratively justified and free of franchise cues.

When privacy protection is requested, choose one of these separately from L1–L3:

• Recognizable stylization: retain identity but render all pixels in a coherent illustrated or painterly language.
• Identity-softened silhouette: retain pose and broad clothing silhouette while simplifying facial detail. Use only when the user prioritizes privacy over likeness.

Build the transformation card

Before generating, write a compact internal card:

Hero fact: who/what the visible person is in this image
Identity locks: face, hair, pose, hands, proportions, accessories
Source anchors: 3-5 scene or clothing facts that must survive
Transformation level: L1 / L2 / L3
Narrative verb: depart / endure / cross / return / guard / discover
Epic premise: one sentence, grounded in the visible source
Scale device: one dominant monumental relationship
World: location, era cues, weather, horizon, foreground, background
Light and palette: key direction, dominant/support/accent colors
Protected zones: face, hands, meaningful object, readable evidence
Forbidden drift: exact changes that would make the result false or generic
Comparison topology: diptych / match-cut seam / reveal window

Do not generate until the card can explain why this specific person belongs in this specific world.

Compose the epic world

Read references/visual-directions.md when choosing the visual world.

Establish one scale relationship

Use one primary device, not five competing spectacles:

• a lone person against a storm wall and distant fleet;
• a human figure at the threshold of colossal ruins;
• a traveler crossing a ridge above an immense sea;
• a returning figure facing a city at dawn;
• a guardian beside a vast archive, map, bridge, or harbor.

The hero must remain the semantic center even when not the largest object. Use gaze, light, directional lines, and atmospheric depth to connect the world to the person.

Use heroic light without plastic glamour

• Shape the face with believable directional light that matches the new sky.
• Prefer restrained rim light, weather light, reflected sea light, or dawn backlight.
• Preserve skin texture and age. Avoid beauty-filter skin, waxy highlights, body reshaping, and artificial eye enlargement.
• Use a restrained palette: one dominant atmosphere, one supporting earth/stone tone, and one controlled warm or cool accent.

Preserve believable anatomy and contact

• Keep hands, fingers, feet, chair contact, desk contact, and carried objects anatomically plausible.
• If the source is seated, do not force a standing hero pose. Turn the seated posture into a command, navigation, archive, or watchkeeping scene.
• If the source is cropped, do not invent missing limbs unless the requested output requires expansion and the reconstruction passes inspection.

Generate in two passes when needed

Prefer a controlled two-pass workflow over one overloaded prompt:

1. World pass: preserve the locked person and establish environment, scale, weather, and light.
2. Character pass: only if L2/L3 is selected, refine clothing or props while keeping face, pose, proportions, and source anchors unchanged.

Every edit prompt must explicitly include:

• SOURCE PERSON IS LOCKED followed by identity locks;
• the transformation level and permitted changes;
• the single scale relationship and narrative verb;
• exact world layers from foreground to horizon;
• light direction and restrained palette;
• protected zones and forbidden drift;
• bans on copied film/franchise language, extra people, text, logos, watermarks, fantasy clutter, face replacement, anatomy changes, and glamor retouching.

Do not ask the image model to typeset Chinese titles. Generate the clean image first.

Inspect and revise

View the generated image at full size and thumbnail size. Reject or revise if any condition fails:

• The person is not immediately recognizable as the source person.
• Face geometry, age, hair, skin tone, body proportions, pose, gaze, hands, or camera angle drifted.
• The lighting on the person does not belong to the new world.
• The background looks like a generic game splash screen rather than a coherent physical place.
• Spectacle overwhelms the person or lacks a readable narrative direction.
• New props, symbols, people, text, or logos appeared without permission.
• The result resembles a specific protected film campaign or actor.
• The result relies on orange-and-teal grading, sparks, debris, glow, or gold armor alone to signal “epic.”

Revise the smallest failed region. Do not regenerate an acceptable face merely to fix a distant cloud.

Deliver the contrast

Keep three separate files when the user wants a topic-ready result:

1. clean transformed image without text;
2. original-versus-result comparison image;
3. optional cover/poster with locally typeset title.

Choose one comparison topology based on source geometry:

• Diptych: original occupies 30–40%, transformation occupies 60–70%; best for a clear result reveal.
• Match-cut seam: align the same face/body across a vertical or diagonal seam; best when pose and crop match closely.
• Reveal window: preserve the original as a smaller evidence window over the epic result; best when aspect ratios differ.

Never label the source “失败” or the result “完美.” Prefer factual labels such as 原始照片 / 世界扩展 or BEFORE / EPIC PASS. Keep the person's face large enough to compare on a phone.

For a #ai史诗大片 tutorial or result post, also provide:

• one-sentence premise;
• transformation level;
• 3–5 identity locks;
• one honest limitation or failed detail;
• a short reusable prompt skeleton without private information.

Final quality gate

• Does the transformation depend on this source person and pose rather than a replaceable fantasy hero?
• Is the hero readable first, followed by the monumental world and then the narrative detail?
• Is there one dominant scale device and one coherent light source?
• Can the original and result be compared quickly at mobile size?
• Are identity, privacy, anatomy, source evidence, and user intent protected?
• Is the image original and free of copied film/franchise assets, branding, and typographic imitation?
• Are the clean image, comparison, and editable/local title layer kept separate?