Lifescape Outfit Image Generation (Xiaohongshu Style)
Each time, clearly display the product on the model specified by the user, making the photo feel like a moment captured by her or a friend in a real setting. The character's identity is determined by the current face image and remains consistent throughout a set of images for the same model. Multi-model tasks bind to their respective references. When the user only provides face and product images, autonomously select a well-reasoned plan and generate images as requested. Default to a phone-recording feel and natural ambient light. Prioritize user-specified style, references, quantity, aspect ratio, and delivery method. When only a plan, prompt, or review is requested, deliver that content only.
First Determine Available Materials and Reference Roles
First review the actually accessible materials for this task, and identify the purpose of each image based on user annotations and task recognition. A single image can serve multiple explicitly specified purposes. When multiple products or color versions are presented and the goal is unclear, first confirm which specific item to showcase. When reusing materials across conversations, rely on the currently readable original images.
- Face images constrain recognizable identity, facial feature relationships, and personal characteristics. Select face references suitable for the shooting angle; expressions change with the character's state.
- Use body references according to the user's purpose. To reproduce the model's actual body proportions, use the user-specified full-body image as the basis. When the user provides an alternative proportion reference, the face image determines identity, and the body image guides the expected head-to-body ratio, shoulder-waist-hip contours, and torso-leg proportions, treating it as a virtual body setup. When changing poses and shoe types, consider perspective, weight-bearing, garment shaping, and the impact of high heels on visual proportions. Keep uncertain bone positions and precise dimensions that a single photo cannot confirm. When only a face image is provided and no body shape is specified, establish a natural virtual body and provide a brief description. Fix one body description per group, detailing head size relative to shoulder width, torso-leg length relationship, and shoulder-hip contours; reuse it for subsequent images. Prioritize features supported by visible references, only naturally completing occluded areas.
- Product images constrain recognizable colors, weave, silhouette, length, placket, pockets, and other structures. First, identify the primary product and any user-specified matching items; treat other items in the image as optional matching candidates. Complete any missing parts (e.g., pant legs, socks, shoes not shown in the screenshot) through styling design and explain the choices in the plan. Show perspectives of the primary product that have evidence. If the user requests a part lacking evidence, ask for additional images. When multiple perspectives conflict, first verify they are the same model and color.
- Style references guide color schemes, ambient lighting, shooting distance, composition, and character state. Face and product images constrain identity and clothing respectively.
- User-provided expression or pose references can come from other people; extract gaze direction, expression intensity, action relationships, and shooting angle as per the specified use. The character's identity is still determined by the current face image; body and product follow their respective references.
Each task uses the body reference selected for this task. When the user provides a full-body proportion reference showing the clothing being worn, base judgments only on the visible overall silhouette. Keep waist-hip positions, knee positions, and true leg length under loose clothes and long skirts as uncertain.
When references conflict, follow the user's explicitly stated purpose and version for this task. If style, action, or props obstruct product display, first adjust the outfit and composition. Product color should be based on the specified reference; brightness and color shifts in the final image must be explainable by ambient light.
Clarify the Relationship Between Multiple Images
First determine if the user wants one set of coordinated images, different styling options, or independent multi-scene materials. If unspecified, default to organizing them as one set and explain this assumption in a brief plan.
Within a set, maintain consistency in the character, body, product, and after completing the outfit: agreed hairstyle, shoes, bag, and environmental continuity. Create variation through shot scale, action, and display focus. For multi-scene or outfit-change tasks, allow corresponding changes as needed. Before generation, briefly confirm the complete outfit for the set, fixed elements, and each image's role; reuse these conventions in subsequent calls. Each image should add readable product information or a meaningful change in the character's lifestyle state.
Except for parts the user intentionally wants unified, compare each image's body support, orientation, action phase, gaze, expression intensity, and framing. For repetitive scenes, choose substantial variations suitable for the product and location, then coordinate other details around them. Changing only hand positions under the same stance usually adds limited information.
Derive Photos from Product and Purpose
- First determine what the audience needs to see. Identify the features of the product worth showcasing and select the primary task for this image. Consider occlusion by hair, props, layering, and poses on key structures. The set can separately communicate overall proportion, drape, and detail; choose aspect ratio and framing based on the display task.
- Complete the full outfit first, then design the shooting method. Prioritize user-specified style. Otherwise, decide the styling direction based on the product's silhouette, material, color, and wearing context. Choose tops or layering, bottoms, socks/shoes, and necessary accessories. Clarify the connection between clothing length and waistline, trouser/skirt length and shoe opening, and whether shoe toe, sole, and garment volume are harmonious. Briefly inform the user of these choices and actually write them into the image generation input. Select a daily-life location compatible with the outfit, then decide background brightness, ambient light, distance, and angle. Style terms like INS, rich-girl, or Japanese must be translated into specific items and proportions.
- Use lifestyle actions to link body and expression. Determine what the character is doing, where their attention is focused, and which phase of the action is captured. Choose well-founded expression intensity and visible cues, ensuring that the eyes, mouth, head posture, and body tension/tone are coordinated. First determine the body's support and orientation, then arrange hands, feet, and fabric. Natural standing, sitting/leaning, walking/stopping, adjusting clothes, and interacting are all viable; choose based on the clothing and environment.
When automatic styling, complementing product images' missing worn parts, or correcting the look is needed, read Complete Outfit and Footwear Selection. Complete a viable outfit first, then showcase it with expression and pose.
When planning expressions, poses, and set variations, read Expression and Pose Design. Use the selection methods and action examples to expand the plan, combining them for the current character, product, and shooting task.
When referencing blogger poses or needing to expand combinations, read Blogger Cover Pose Deconstruction. Transfer body support, head-body orientation, hand division of labor, and composition relationships to the current character and product, re-establishing contact and weight-bearing for the current scene.
Implement the Plan into Generation Input
Briefly state the specific outfit for the set, the relationship between images, each image's display goal, and any necessary assumptions, then execute as requested. Only ask the user to supplement if missing information would change the character, product, or explicitly requested visuals and cannot be reasonably completed.
Each generation pass the applicable original face images, product images, and body shape/style references as actual image inputs to the tool, annotating each reference's purpose. When retouching, provide the image to be retouched along with relevant original references. Adapt to the tool's accepted reference formats and quantity; if necessary references cannot be passed or the tool is unavailable, explain the specific limitations and what has been completed.
Each image generation input must clearly state the primary product and any matching items visible or likely to appear in the frame. Use specific visible descriptions for shoe types and trouser/skirt silhouettes in the complete set. When a person appears in the frame, actually write the visible body relationships, camera position, and main support points for that shot scale into the generation input. For full-body shots, when the user hasn't specified the shooting method, default to stepping back, a phone main-camera perspective, lens approximately at waist-to-chest height and nearly horizontal, with room above the head and below the feet. To elongate legs, achieve this through outfit and pose, maintaining the agreed torso-leg length relationship.
Generate each image separately. If the user requests a collage, execute according to their layout. When a set includes full-body images, first generate one full-body image showing head to toe with clear support; after confirming the outfit and proportions, expand to other images. If the user only requests half-body or close-ups, verify within that shot scale. For subsequent generations, continue passing applicable original references; may attach the first approved full-body image as a supplementary body reference for the same set, noting its purpose.
Verify Images After Generation
Check each image against its display goal. Cross-check the visible identity, body shape, and product against the original references. Then verify that the complete outfit reflects the chosen style, particularly examining the waist-hip layered look, trouser/skirt hem shape, shoe opening connection, and shoe volume. For standing and sitting poses, check knee and hem weight-bearing, pointing out abnormal bunching and overall proportion imbalance. Then check expression, gaze, weight-bearing, grip, ambient light, and material. For expression, focus on whether the eyes and mouth together convey the current state. For pose, focus on whether body support and action are realistic. For the set, additionally check agreed consistency and actual visible differences; determine from the resulting images whether expression and pose variations are realized. Close-ups can only verify visible clothing parts. Mark obscured or low-resolution content as "unverified". For completely covered faces, only check visible cues.
For full-body images, check the head-to-shoulder size relationship, shoulder-to-hip and hip-to-foot length relationship, hip-knee-ankle connection, and both feet relative to the ground. For half-body images, check visible head-shoulder, neck, and arm relationships. If the head is significantly small, shoulder width deviates notably from the setting, torso is compressed, legs are elongated, or joints are misaligned, mark as "needs correction". If the first full-body image fails, correct it first. Combine pose, heel height, and perspective; clothing waistlines or skirt hems are only worn cues.
For mirror selfies, intentionally covered faces, plan comparison, reviews, partial retouching, or when encountering issues like disproportional body, template face, staged feel, floating, or material deviation, read Realism Judgment and Correction.
First correct deviations affecting the character, product, or display purpose, then correct local naturalness. Use targeted edits for local defects; for overall composition or structural errors, return to the original reference to adjust the generation plan. After each round of correction, re-verify key constraints. If continuous correction shows no improvement or begins to damage already correct parts, stop adding modifications, keep the better version, and explain any remaining unaddressed evidence or trade-offs. Deliver the selected images, clearly stating any unresolved deviations with actual impact; mark versions with key deviations as "needs correction".