wzj177/ecommerce-image-suite

面向电商商品图生产场景的图片套图 Skill。仓库提供一套从商品图分析、卖点提炼、Prompt 组织到成图输出的脚本与配置，适合用在淘宝、京东、拼多多、抖音等国内平台，以及独立站、Amazon 等跨境场景。

What it solves

This project provides a streamlined workflow for creating professional e-commerce product image sets. It automates the process of transforming a single product photo into a structured set of marketing images tailored for platforms like Amazon, Taobao, JD.com, and TikTok, reducing the manual effort required to generate various required image types (e.g., white background, lifestyle, and detail shots).

How it works

The system uses a multi-step pipeline:

  1. Visual Analysis: A script (analyze.py) analyzes the original product image to extract structured product information.
  2. Prompt Organization: The tool organizes selling points and prompts based on the platform and image type requirements.
  3. Image Generation: A script (generate.py) uses AI image generation providers (such as Alibaba Tongyi, ByteDance Doubao, OpenAI, Google Gemini, or Stability AI) to produce the final image set.
  4. Optional Video Generation: A script (generate_video.py) can create videos based on the generated image sets.

Who it’s for

It is designed for e-commerce sellers and marketers operating on domestic Chinese platforms or international cross-border sites who need to produce high-quality, standardized product imagery quickly.

Highlights

  • Comprehensive Image Types: Supports generating white-background main images, core selling point images, material detail shots, lifestyle scene images, model shots, and multi-angle collages.
  • Multi-Provider Support: Compatible with several major AI image generation APIs including Tongyi, Doubao, OpenAI, Gemini, and Stability AI.
  • Structured Workflow: Converts raw images into structured JSON data to ensure consistency across images, copywriting, and potential video generation.
  • Ready-to-use Examples: Includes test cases for various categories like apparel and electronics.

Related

  • Project
    Shilin-LU/MACE

    MACE is a finetuning framework for diffusion models that enables the mass erasure of up to 100 unwanted concepts, such as celebrities or explicit content, while preserving unrelated model knowledge.

  • Project
    popopo-99/zy-cinematic-realism

    A cinematic visual workflow for LLMs that transforms narrative ideas into professional, model-specific prompts for AI image generators to achieve authentic cinematic realism.

  • Project
    worldbench/DiffusionOPSD

    DiffusionOPSD is an open‑source implementation of the “On‑Policy Self‑Distillation” algorithm for fine‑tuning diffusion image generators (Stable Diffusion 3.5‑M and Z‑Image‑Turbo). It repeatedly collects low‑noise states from a frozen behavior policy, builds explicit positive/negative targets using reward‑gradient steps, trains a new policy to match those targets, and refreshes the behavior policy via EMA. The repo provides scripts for installing dependencies, downloading reward‑model checkpoints, running quick smoke tests, and launching full‑scale distributed training for any of seven public evaluators (PickScore, CLIPScore, HPSv2.1/v3, Aesthetic, ImageReward, DeQA) or mixed‑reward objectives. Reported results show higher held‑out scores on 19/20 settings and 40‑63 % less GPU‑time versus prior methods. Licensed under Apache 2.0.

  • Project
    facefusion/facefusion-pinokio

    FaceFusion Pinokio is a face manipulation platform that allows users to swap and modify faces in images and videos with a simplified installation process via Pinokio.