P2M — Reference-Locked Model Swap (Nano Banana Edition)

Image 2 May Contain a Model — Handle Correctly

Image 2 is typically a standard e-commerce product shot. This often means the garment is photographed on a model (white or neutral background, front-facing pose) — not a hanger shot or flat lay.

This creates a specific risk: Nano Banana sees two different people across Image 1 and Image 2, and may try to blend their facial features, body type, or skin tone into the generated frame.

The prompt must explicitly instruct Nano Banana to:

1. Extract clothing information only from Image 2 — color, texture, construction.
2. Completely disregard the person in Image 2 — their face, hair, body, and skin tone are irrelevant.
3. Use only the person in Image 1 as the model for the generated photo.

This is the most critical instruction in the entire prompt. It must be stated at the top, before any other content, and repeated in the clothing lock section.

Image 1 Shows the Old Garment — Handle Correctly

Image 1 is the completed model-scene photo and will usually show the model wearing some garment. That garment is not the target — it is being replaced.

The prompt must explicitly instruct Nano Banana to:

1. Extract model identity, body, pose, camera, and scene only from Image 1.
2. Completely replace the garment in Image 1 with the garment described by Image 2.
3. Never blend the two garments — the original outfit in Image 1 is gone.

Why Two Images Are Better Than One

When only the product image is used, the model, pose, scene, and framing have to be invented — losing the specific look the user already validated. When only the model photo is used, the garment's color is already filtered by the scene's lighting and camera, so Nano Banana may reproduce a distorted color. By providing both:

• Image 1 is absolute authority for the model and scene. The user's existing shoot (face, body, proportions, framing, location, mood) is preserved 1:1.
• Image 2 is absolute authority for the garment. Color and construction are read from an unfiltered source, so the new outfit appears true-to-product.

How to Label the Two Images in Your Message to Nano Banana

When you submit to Nano Banana, label the images explicitly in your prompt opening — and immediately state both the model-replication rule and the Image 2 person-exclusion rule:

I am providing two reference images:
- Image 1 (Model + Scene + Framing Reference): a completed fashion photo.
  Use this as the ONLY authority for the person, their body, proportions,
  pose, camera angle, and framing, and for the scene environment. The
  garment currently worn in Image 1 is being REPLACED and must not appear
  in the output.
- Image 2 (Garment Reference ONLY): a product photo of the new garment.
  Extract ONLY the clothing details from this image — exact color, fabric
  texture, and construction. Completely ignore any person shown in Image 2.
  Their face, hair, body, and skin tone are irrelevant.

Input Requirements

This skill requires exactly two images plus an optional brief text description:

• Image 1 — Completed model-scene photo. May be full body, 3/4, waist-up, chest-up, or head-and-shoulders. Whatever crop it is, that crop must be preserved in the output.
• Image 2 — Product photo of the garment. Can be on a model (white/neutral background), on a hanger, flat lay, or mannequin. Front view preferred for clearest construction detail.
• Optional text — Any additional intent not clearly readable from the images (e.g., "slightly warmer late-afternoon light" or "soften the background a bit more").

How Nano Banana Handles Dual-Image Input

Nano Banana (Gemini Flash Image) is a native multimodal model — it processes both images simultaneously using its unified reasoning architecture. This means:

1. It reads both images before generating. Specify the role of each image explicitly so the model knows which to treat as model+scene authority vs. clothing authority.
2. Identity Lock is native. Phrasing like "Using the exact person from Image 1, maintain identical facial structure, hairstyle, skin tone, body proportions, and pose" activates built-in character consistency using Image 1 as the anchor.
3. Product Fidelity Lock is explicit. The phrase "Reproduce the clothing from Image 2 with 100% color and texture accuracy — Image 2 overrides any lighting-induced color variation in Image 1" tells Nano Banana to prioritize the product photo for garment truth.
4. Natural language is preferred. Write the prompt as clear, structured prose — not comma-separated keyword dumps.
5. Semantic positives only. Never use "no X" — describe what you DO want instead.
6. Framing must be stated explicitly. Nano Banana will otherwise default to a full-body shot; if Image 1 is waist-up, say "framed from waist up — the crop must match Image 1 exactly".

Instructions

Step 1: Build the Four Lock Blocks

Before writing the final prompt, extract and write out four lock blocks. Always note the image source for each piece of information.

IDENTITY LOCK — sourced from Image 1 (model scene photo):

• Face: key structural features (jaw, eye shape, nose bridge, distinguishing marks).
• Hair: exact color, length, style, part direction, loose/tied.
• Skin tone: specific description (e.g., "warm ivory", "golden olive", "deep espresso").
• Makeup: level and style (e.g., "minimal, natural brow, neutral lip").
• Accessories: list every visible item exactly — earrings, necklace, rings, watch, glasses — or state "none visible".

BODY & POSE LOCK — sourced from Image 1:

• Body type: overall build (slim, athletic, curvy, plus, etc.) — describe what you actually see; do not impose any body standard.
• Proportions: shoulder-to-waist-to-hip ratio, limb length, height-to-head ratio.
• Exact pose: standing/sitting/leaning/walking, weight distribution, torso rotation, head tilt.
• Hand placement: exact position of each hand (in pocket, at side, touching hair, holding nothing, etc.). If the reference shows hands holding a prop, keep or drop the prop as the user prefers — default: keep.
• Facial expression: smiling/neutral/serious/laughing, eye direction.

FRAMING & CAMERA LOCK — sourced from Image 1:

• Shot type: name it exactly — Extreme Long Shot (ELS), Long Shot (LS), Medium Long Shot (MLS, knees-to-head), Medium Shot (MS, waist-up), Medium Close-Up (MCU, chest-up), Close-Up (CU, head-and-shoulders), or Extreme Close-Up (ECU).
• Camera angle: eye level, low angle, high angle, 3/4, profile, front, back, degree of rotation.
• Focal length feel: wide (24–35mm environmental), normal (50mm), portrait (85mm), telephoto (100mm+).
• Crop boundaries: state where the frame begins and ends — top of head, mid-thigh, knees, ankles, etc.
• PARTIAL-BODY REPLICATION RULE: If Image 1 is not full-body, the output must use the identical crop. Do not "helpfully" extend the frame to show the whole body. If Image 1 is waist-up, the output is waist-up. If Image 1 is head-and-shoulders, the output is head-and-shoulders. State this explicitly in the final prompt.

CLOTHING LOCK — sourced from Image 2 (product photo); replaces the garment currently shown in Image 1:

• Color: read from Image 2 — state the precise name and note "Image 2 is color authority".
• Pattern/print: motif, placement, scale, orientation — or "solid" — from Image 2.
• Silhouette: fit category + length — from Image 2.
• Construction: collar type, neckline, sleeve style, closure type and exact count, pocket placement, hem finish, hardware, logos — from Image 2.
• Fabric texture: matte/sheen/ribbed/woven/knit — from Image 2 (unaffected by scene lighting).
• Replacement note: explicitly state that the garment currently worn in Image 1 is being replaced and must not appear in the output.

Step 2: Build the Scene Variation Block

The user wants the scene from Image 1 preserved — but they have granted bounded creative freedom to subtly reinterpret lighting and depth of field. Describe exactly what can and cannot change.

What is allowed to vary (mildly):

• Lighting direction — may shift a few degrees.
• Color temperature — may drift slightly warmer or cooler.
• Light intensity — may soften or strengthen.
• Background depth of field — may be softened with gentle bokeh.
• Atmospheric mood — a touch more haze, a softer glow, a gentler hour.

What must NOT change:

• The model (face, body, skin tone, hair, pose, expression).
• The framing / crop / camera angle.
• The location and its structural elements (walls, furniture, windows, floor, key background objects).
• The overall color palette and grading identity of the scene.
• The type of light source (window light stays window light; studio strobe stays studio strobe).

This is the operational meaning of the user's "略微修改场景的光景，虚化，适当发挥一些想象" requirement.

Step 3: Write the Nano Banana Prompt

Structure the final prompt in this exact order:

1. [IMAGE ROLE DECLARATION] — Label Image 1 and Image 2 and their respective authorities
2. [ABSOLUTE MODEL REPLICATION RULE] — The only person in the output is the Image 1 model
3. [IMAGE 2 PERSON EXCLUSION RULE] — Ignore any person shown in Image 2
4. [GARMENT REPLACEMENT RULE] — The garment in Image 1 is replaced by the garment in Image 2
5. [IDENTITY + BODY + POSE PROSE] — from Image 1
6. [FRAMING & CAMERA PROSE] — from Image 1, including the partial-body replication rule
7. [CLOTHING PROSE] — from Image 2, with Image 2 as color authority
8. [SCENE VARIATION PROSE] — what may vary, what must not
9. [TECHNICAL REQUIREMENTS] — photorealism, resolution, quality

Mandatory Constraints

Dual-Image Reference (NON-NEGOTIABLE)

• Always declare both images at the top of the prompt with their roles.
• Image 1 overrides Image 2 for person, body, pose, camera, and framing — state this explicitly.
• Image 2 overrides Image 1 for clothing color, texture, and construction — state this explicitly.
• The garment in Image 1 is replaced and must not appear in the output — state this explicitly.

Model Replication (NON-NEGOTIABLE)

• Face, hair, skin tone, makeup, and accessories must be 1:1 with Image 1 — no drift.
• Body type, proportions, and height-to-head ratio must be 1:1 with Image 1 — do not slim, thicken, lengthen, or stylize the body in any direction.
• Pose, hand placement, weight distribution, head tilt, and expression must be 1:1 with Image 1.
• Activate with: "Using the exact person from Image 1, maintain identical facial structure, bone structure, interpupillary distance, hairstyle, skin tone, body type, body proportions, pose, hand placement, and expression."

Framing Replication (NON-NEGOTIABLE)

• Camera angle, shot type, focal-length feel, and crop boundaries must be 1:1 with Image 1.
• If Image 1 is partial-body (waist-up, chest-up, head-and-shoulders, etc.), the output must use the identical crop. Do not extend the frame.
• State explicitly: "The output crop must match Image 1 exactly — if Image 1 is framed from [waist up / chest up / head-and-shoulders], the output is framed from [the same]. Do not show body parts that are not visible in Image 1."

Product Fidelity (NON-NEGOTIABLE)

• Exact color from Image 2 — state "Image 2 is the color authority; the true garment color is as shown in Image 2; maintain this color truth even as scene lighting varies".
• Exact construction detail count (buttons, pockets, seams) from Image 2.
• Exact fabric texture from Image 2.
• Never invent details not visible in Image 2.

Scene Variation (BOUNDED)

• Only the specific variations listed in the SCENE VARIATION block are permitted.
• Location, structural elements, furniture, walls, windows, and overall scene identity must remain the same as Image 1.

Output Format

Always output in this structure:

IMAGE ROLE ASSIGNMENT

• Image 1 (Model + Scene + Framing Reference): [brief description of what's in the model scene photo]
• Image 2 (Garment Reference ONLY): [brief description of what's in the product photo]
• Image 1 is authority for: person (face, body, skin tone, proportions, pose, expression), camera (angle, shot type, crop), scene (location, base lighting)
• Image 2 is authority for: clothing (color, fabric texture, construction)

IDENTITY LOCK (sourced from Image 1)

• Face: [structural features]
• Hair: [color, length, style, part]
• Skin tone: [specific description]
• Makeup: [level and style]
• Accessories: [exact list or "none visible"]

BODY & POSE LOCK (sourced from Image 1)

• Body type: [describe as seen — no imposed standard]
• Proportions: [shoulder-waist-hip, limb length, head ratio]
• Pose: [standing/sitting/etc., weight distribution, rotation, head tilt]
• Hand placement: [exact position of each hand]
• Expression: [facial expression, eye direction]

FRAMING & CAMERA LOCK (sourced from Image 1)

• Shot type: [ELS / LS / MLS / MS / MCU / CU / ECU — name it]
• Camera angle: [eye level / low / high / 3/4 / profile / front / back]
• Focal length feel: [wide / normal / portrait / telephoto]
• Crop boundaries: [frame starts at ___, ends at ___]
• Partial-body replication: [if partial, state exactly what is and is not in frame; the output must match this 1:1]

CLOTHING LOCK (sourced from Image 2 — replaces the garment in Image 1)

• Color: [precise name] — Image 2 color authority
• Pattern: [description or "solid"] — from Image 2
• Silhouette: [fit + length] — from Image 2
• Construction: [all visible details, exact counts] — from Image 2
• Fabric: [texture and material] — from Image 2
• Replaces: [one-line note that the garment currently worn in Image 1 is being replaced]

SCENE VARIATION (bounded reinterpretation of Image 1's scene)

• Allowed to shift: [list the specific lighting / bokeh / mood tweaks you are permitting]
• Must remain: [location, structural elements, furniture, overall grading identity, type of light source]

NANO BANANA PROMPT

I am providing two reference images:
- Image 1 (Model + Scene + Framing Reference): [describe Image 1 — the completed
  model scene photo, including shot type and crop, e.g. "waist-up of a woman by
  a café window in warm afternoon side light"]
- Image 2 (Garment Reference ONLY): [describe Image 2 — e.g. "a forest-green
  cable-knit cardigan, flat lay on neutral background"]

CRITICAL — IMAGE 2 PERSON EXCLUSION RULE:
Image 2 may show a model wearing the product. Completely ignore that person.
Do not reference, blend, or incorporate any feature of the Image 2 model — not
their face, hair, skin tone, body type, or pose — into the generated output.
Extract from Image 2 ONLY: garment color, fabric texture, and construction.

CRITICAL — MODEL REPLICATION RULE:
The only person in the output is the exact person from Image 1. Use Image 1
as the sole authority for facial structure, bone structure, interpupillary
distance, hairstyle, skin tone, makeup, accessories, body type, body
proportions, pose, hand placement, weight distribution, head tilt, and
expression. No facial drift. No body reshaping. No pose change.

CRITICAL — GARMENT REPLACEMENT RULE:
The garment currently worn by the model in Image 1 is being replaced in full
by the garment shown in Image 2. The original outfit in Image 1 must not
appear in any form in the generated output. The new garment is the one
described in Image 2.

CRITICAL — FRAMING REPLICATION RULE:
The output crop must match Image 1 exactly. [If Image 1 is waist-up, the
output is waist-up. If head-and-shoulders, the output is head-and-shoulders.
State the specific crop.] Do not extend the frame to show body parts that
are not visible in Image 1.

MODEL (identity + body + pose anchor — Image 1):
[IDENTITY LOCK + BODY & POSE LOCK content as prose]

CAMERA & FRAMING (Image 1):
[FRAMING & CAMERA LOCK content as prose, repeating the exact crop]

OUTFIT (color and texture anchor — Image 2; replaces the garment in Image 1):
[CLOTHING LOCK content as prose, explicitly noting Image 2 as color authority]
The true garment color and fabric texture are as shown in Image 2. Maintain
this color truth even as the scene lighting varies.

SCENE (Image 1 environment, with bounded reinterpretation):
[Describe the Image 1 scene as preserved. Then describe the permitted
variations — slight lighting shift, gentle bokeh, softer glow, etc. Then
state what must remain unchanged.]

TECHNICAL REQUIREMENTS:
Photorealistic fashion photography. Cinematic color grading matching the
identity of Image 1. Accurate focal length and depth of field consistent with
the shot type. 8K resolution. Masterpiece quality fashion editorial photography.

Example

Input:

Image 1: A waist-up photo of a woman sitting by a café window in late-afternoon
         side light, wearing a plain white t-shirt. Warm cinematic grading.
         Relaxed, gentle smile, looking slightly off-camera to the right.
Image 2: A forest-green cable-knit cardigan, flat lay on a neutral wood
         background, with visible chunky cable pattern, wooden buttons, and
         ribbed cuffs.

Output:

IMAGE ROLE ASSIGNMENT

• Image 1 (Model + Scene + Framing Reference): Waist-up portrait of a woman at a café window, late-afternoon warm side light, currently wearing a white tee — this is the only person who appears in the output and the crop is waist-up.
• Image 2 (Garment Reference ONLY): Forest-green cable-knit cardigan, flat lay — extract garment details only; any person implied in Image 2 is irrelevant.
• Image 1 is authority for: face, body, skin tone, proportions, pose, expression, camera angle, waist-up crop, café window scene, warm side light.
• Image 2 is authority for: forest-green color, chunky cable-knit texture, wooden-button construction, ribbed cuffs.

IDENTITY LOCK (sourced from Image 1)

• Face: Soft oval jaw, almond-shaped eyes, straight nose bridge, light natural brows.
• Hair: Dark chestnut brown, shoulder length, loose, middle part, falling naturally in front of the shoulders.
• Skin tone: Warm ivory with a light sun-kissed undertone.
• Makeup: Minimal — natural brow, neutral rose lip, subtle mascara.
• Accessories: Small gold stud earrings on both ears, thin gold chain necklace. No rings, no watch visible in the waist-up crop.

BODY & POSE LOCK (sourced from Image 1)

• Body type: Slim-to-average build as seen in Image 1 — reproduce exactly, do not alter.
• Proportions: Narrow shoulders relative to head, natural waist, proportional torso length for a waist-up crop.
• Pose: Seated, slight forward lean, torso rotated about 15° toward camera-left, head tilted a few degrees to the right.
• Hand placement: Right hand rests lightly on the café table in the lower edge of the frame; left hand not visible in the waist-up crop.
• Expression: Relaxed, gentle closed-mouth smile, eyes looking slightly off-camera to the right.

FRAMING & CAMERA LOCK (sourced from Image 1)

• Shot type: Medium Shot (MS), waist-up.
• Camera angle: Eye level, front with slight 3/4 rotation toward camera-left.
• Focal length feel: 85mm portrait — shallow but not macro depth of field, clean subject separation.
• Crop boundaries: Frame begins just above the top of the head, ends at the mid-waist of the torso just above the table edge.
• Partial-body replication: Image 1 is waist-up — the output is waist-up. The hips, legs, and feet are NOT in frame and must NOT be added to the output. Do not extend the crop downward to show a full body.

CLOTHING LOCK (sourced from Image 2 — replaces the white tee in Image 1)

• Color: Forest green — Image 2 color authority — maintain across the output even though Image 1's warm light may tint it.
• Pattern: Chunky vertical cable-knit pattern across the front panels and sleeves — from Image 2.
• Silhouette: Relaxed-fit cardigan, open-front style worn closed at the chest, length extends below the waist (out of frame in this waist-up crop).
• Construction: Five wooden buttons down the center front (only the top three are in frame for a waist-up crop), shawl or notched collar as shown in Image 2, ribbed cuffs on the long sleeves, ribbed hem (out of frame).
• Fabric: Heavy wool-blend cable knit with visible fiber texture, matte surface — from Image 2.
• Replaces: The plain white t-shirt currently worn by the model in Image 1 is replaced in full by this cardigan and must not appear in the output.

SCENE VARIATION (bounded reinterpretation of Image 1's scene)

• Allowed to shift: Late-afternoon side light may drift very slightly warmer and softer, as if a touch later in the hour. Background (café interior through the window) may have slightly softer bokeh. Atmospheric haze may be a hair more pronounced.
• Must remain: The café window, the table edge, the visible interior elements, the overall warm cinematic grading, the direction of the key light (from camera-left), and the general mood of a quiet afternoon café moment.

NANO BANANA PROMPT

I am providing two reference images:
- Image 1 (Model + Scene + Framing Reference): A waist-up portrait of a woman
  sitting by a café window in late-afternoon warm side light, currently
  wearing a plain white t-shirt. Warm cinematic grading, relaxed gentle smile.
- Image 2 (Garment Reference ONLY): A forest-green cable-knit cardigan, flat
  lay on a neutral wood background, with chunky vertical cable pattern,
  wooden buttons, and ribbed cuffs.

CRITICAL — IMAGE 2 PERSON EXCLUSION RULE:
Image 2 shows the cardigan as a flat lay. Do not reference, blend, or
incorporate any person or human figure implied by Image 2 into the output.
Extract from Image 2 ONLY: forest-green color, cable-knit texture, wooden
button count and placement, ribbed cuff construction.

CRITICAL — MODEL REPLICATION RULE:
The only person in the output is the exact woman from Image 1. Use Image 1
as the sole authority for facial structure, bone structure, interpupillary
distance, dark chestnut shoulder-length hair with middle part, warm ivory
skin tone, minimal natural makeup, small gold stud earrings, thin gold chain
necklace, slim-to-average build, narrow shoulders, natural proportions,
seated forward-lean pose with torso rotated about 15° toward camera-left,
head tilted a few degrees to the right, right hand resting on the table
edge, relaxed gentle closed-mouth smile, eyes looking slightly off-camera
to the right. No facial drift. No body reshaping. No pose change.

CRITICAL — GARMENT REPLACEMENT RULE:
The plain white t-shirt currently worn by the model in Image 1 is being
replaced in full by the forest-green cable-knit cardigan shown in Image 2.
The white t-shirt must not appear in any form in the output.

CRITICAL — FRAMING REPLICATION RULE:
The output crop must match Image 1 exactly: framed from just above the top
of the head down to mid-waist just above the table edge. This is a waist-up
Medium Shot at eye level, approximately 85mm portrait focal length, with a
slight 3/4 rotation toward camera-left. Do not extend the frame downward.
The hips, legs, and feet are not in frame and must not be shown.

MODEL (identity + body + pose anchor — Image 1):
A woman with a soft oval jaw, almond-shaped eyes, straight nose bridge, and
light natural brows. Dark chestnut shoulder-length hair, middle part, falling
naturally in front of the shoulders. Warm ivory skin with a light sun-kissed
undertone. Minimal natural makeup — natural brow, neutral rose lip, subtle
mascara. Small gold stud earrings and a thin gold chain necklace. Slim-to-
average build with narrow shoulders and natural proportions. Seated with a
slight forward lean, torso rotated ~15° toward camera-left, head tilted a
few degrees to the right. Right hand rests lightly on the café table at
the lower edge of the frame. Relaxed, gentle closed-mouth smile, eyes
looking slightly off-camera to the right.

CAMERA & FRAMING (Image 1):
Eye-level Medium Shot, waist-up, approximately 85mm portrait focal length,
shallow but not macro depth of field, slight 3/4 front angle toward
camera-left. The frame begins just above the top of the head and ends at
mid-waist just above the café table. Waist-up crop must match Image 1
exactly — no extension below the frame.

OUTFIT (color and texture anchor — Image 2; replaces the white tee in Image 1):
Relaxed-fit forest-green cable-knit cardigan, worn closed at the chest. The
true color is forest green exactly as shown in Image 2 — maintain this color
truth even as the warm afternoon side light falls across it. Chunky vertical
cable-knit pattern across the front panels and visible upper sleeves. Five
wooden buttons run down the center front; only the top three are visible in
the waist-up crop. Ribbed cuffs visible on the long sleeves where they rest
on the table. Heavy wool-blend cable knit with visible fiber texture, matte
surface, exactly as shown in Image 2.

SCENE (Image 1 environment, with bounded reinterpretation):
Same café window and same table edge as Image 1. Same warm cinematic color
grading. Same direction of key light from camera-left. The late-afternoon
light may drift very slightly warmer and softer, as if a touch later in the
hour. The café interior visible through the window may have slightly softer
bokeh. A hair more atmospheric haze is acceptable. The location, window,
table, interior elements, and overall mood remain identical to Image 1.

TECHNICAL REQUIREMENTS:
Photorealistic fashion editorial photography. Cinematic color grading
matching Image 1's warm afternoon identity. 85mm portrait depth of field
with clean subject separation from the softly blurred background. 8K
resolution. Masterpiece quality.

Notes

• Never impose a body standard. Describe the body you actually see in Image 1. Do not slim, thicken, lengthen, or otherwise "flatter" the reference body — the user chose Image 1 specifically for its body.
• Never extend the frame. A waist-up reference stays waist-up. This is the single most common failure mode of image models on this task.
• Never blend the two garments. The Image 1 outfit is gone; the Image 2 outfit is the only clothing.
• Never borrow from the Image 2 person. If Image 2 is shot on a model, that person is invisible to the output.
• Never invent garment details. If a detail is not visible in Image 2, do not describe it. Say "out of frame" or "not visible in Image 2".
• Scene variation is a slider, not a switch. Keep reinterpretation subtle — the user wants the same shoot, not a different shoot.