Moonbox
简介
这是一款老年人AI聊天产品，根据小月与老人的聊天内容，生成老人的人生回忆录。
工具
Midjourney
Runway
Photoshop
Figma
需求
设计产品的UIUX，完善用户体验
小月IP形象设计
IP形象需求
生成一个女孩IP形象，要求亲切的邻家，无攻击性，但又专业可信赖，让老年人有倾诉欲望
工具
Midjourney， 可灵AI
人物风格
粘土3D人物风格
小月卡通
小月真人
Runway生成动画
小月来电啦～
小月在听...
小月正在说...
小月正在生成...
lulula
简介
一款面向25-40岁女性的AI语言互动App
工具
Midjourney
Stable Diffustion
需求
设计产品的UIUX，完善用户体验
AIGC制作所有配图
Lora训练
风格统一
风格：毛玻璃，朦胧，唯美，噪点插画
Midjourney--sref
提示词+风格参考+风格权重
出图效果：颜色鲜艳, 更有艺术感 ,边线模糊
VS
SD-Lora
触发词：lulula
迭代步数：20-30
采样方法： DPM++2M Karras
底模: f.1_dev_fp8
出图效果 ：颜色偏灰, 轮廓更清晰
Reality Booth
简介
一款To B利用AIGC技术赋能电商产品图的服务
工具
Comfy UI
Photoshop
需求
输出完整品牌Identity
使用Comfy UI完成换背景，生成产品图
Comfy UI-产品换背景
1. 精准抠图 & 细节保留
AI 抠图容易导致边缘问题，特别是透明材质。我们用 ControlNet（Depth + Canny），让抠出来的物品保持清晰的轮廓，同时保留原始光影。
2.背景匹配 & 真实感增强
不同产品需要不同的背景，我们会按照客户要求 定制。根据抠出来的产品图融合到定制背景中，并用Lora,controlnet,控制融合效果。
3. 批量处理 & 统一风格
增加批量处理节点，确保大批量生成时风格稳定。
输出图片类型
不同产品，同一背景风格
相同产品，不同角度，同一背景风格
相同产品，在不同光效下，会有颜色变化
产品文字保持清晰不变
输出图片Samples
AIGC-Powered Real Estate Visual Optimization
Background
In the North American real estate market, the Multiple Listing Service (MLS) imposes strict quality standards on property images. We needed to ensure that the 40 property photos we delivered were high-resolution, rich in detail, and visually appealing.
However, traditional image processing methods were time-consuming, costly, and lacked consistency. To overcome these challenges, we explored an AIGC-powered workflow to optimize the entire process.
Goal
Enhanced Image Quality
Increased Efficiency
Empowered Team Collaboration
Challege 01
Low image quality, poor color accuracy.
Since we use a panoramic camera to capture indoor scenes, we first export a panoramic image and then extract 35-40 representative interior shots. For exterior shots, we take around 7 images using mobile devices, which often results in lower resolution. This is especially problematic in overcast conditions or dimly lit indoor environments, where the image quality does not meet the required standards.
Original Images
Solution 01
Photoshop
Image Correction + Camera Raw Filter + Sky Enhancement + Grass Adjustment + Object Removal
Stable Diffusion
Image-to-Image
Model:
老王_Architecutral_MIX V0.5
In the early stages, we conducted extensive research and experiments, testing various architecture and interior design-related checkpoints and LoRAs on Liblib. We evaluated the models' versatility across different scenes and environments, as well as their handling of lighting and shadows. After rigorous testing, we selected Lao Wang's model.
Parameter Settings
Sampling Method: DPM++ 3M SDE Karras (High-quality details)
Steps: 40
CFG Scale (Prompt Guidance Strength): 6
Denoising Strength (Repainting Amount): 0.3
Precise Control with ControlNet
Control Type： Canny
4K High-Resolution Upscaling
Script: Ultimate SD Upscale
This method focuses on overall detail enhancement while maintaining strong global consistency across architectural images.
It provides better light and shadow coherence, effectively avoiding seams or style inconsistencies.
Eliminates visible seams and ensures stylistic coherence
With high efficiency, it is well-suited for batch processing.
Result
Standardized Production Workflow
Efficiency Improvement
45%
To standardize the PS+AIGC workflow and enhance team collaboration, I organized a Company-wide training session for 30+ designers from China and the U.S., ensuring a consistent standardized process for real estate visual optimization.
Hosted AIGC Training Session — Unified the designers' understanding of best practices and efficient workflows.
Created Standardized Workflow Documentation — Detailed instructions covering the installation of Stable Diffusion, plugin configurations, and image optimization methods.
Produced Comprehensive Operating Guides — Ensured designers efficiently produced high-quality visual content following the standardized workflow.
Cross-Team Knowledge Sharing — Promoted efficient collaboration among designers from different regions.
Customer Satisfaction
100%