How AIGC Advances Boost Image Quality: Lessons from Alibaba’s Cyber Project

Introduction

Since July the author has been working on AIGC‑related projects, noticing a clear improvement in generated image quality. The boost comes from rapid developments in large models and open‑source plugins, as well as a deeper grasp of generation routes and parameters.

Cyber Project Overview

Cyber is an internal initiative focused on engineering AIGC, providing end‑to‑end solutions for model training, deployment, testing, and generation workflows. It currently excels in AI mannequin creation and intelligent background generation.

AI Mannequin

The AI mannequin can serve both B2B and B2C scenarios. For B2B, it reduces product‑shoot costs by generating diverse model images from a few clothing photos. For B2C, it enables applications such as AI try‑on rooms and creative cameras.

Technical Solution Types

SD Lora : Suitable for flat clothing images; requires multiple training inputs and may have longer user wait times.

SD Inpainting : Works with real or mannequin‑worn clothing; preserves clothing details accurately.

Midjourney Padding : Generates natural‑looking models for simple colors and styles, though exact clothing details may vary.

Previous‑generation Alchemy Furnace : Open‑source, supports multi‑angle and multi‑form generation.

New‑generation Diffusion (2‑UNet) : Proprietary, not yet open‑source, but research shows superior performance over VITON/tryon GAN.

Effect Overview

Side‑by‑side comparisons show original merchant images versus Cyber‑generated results, demonstrating clearer details, consistent styles, and realistic backgrounds.

Current Experience

Industry Status

The AIGC industry chain includes upstream data annotation and hardware, midstream cloud computing and performance optimization (e.g., Google, OpenAI), and downstream content generation and distribution platforms. Major players have already secured the first wave of revenue from chips and cloud services.

User Acceptance

While public enthusiasm varies, early data suggests AI‑enhanced images can increase click‑through rates in e‑commerce, though concerns about copyright and product‑image fidelity remain.

Social Media Feedback

Positive reactions highlight the novelty and engagement of AI‑generated content, whereas negative comments focus on potential copyright infringement and mismatched product representations.

Front‑end Role in AIGC

Front‑end engineers can lead early technical research, design stable generation workflows, and explore multi‑domain AI generation such as pets, anime, video, and more. They also contribute to building reusable components like background removal, 3D pose editors, image editors, and canvas manipulation tools.

Ideas and Challenges

Copyright‑Driven Creative Stagnation : Over‑reliance on AI‑generated content may hinder original knowledge creation and raise legal restrictions.

Aesthetic Fatigue : Excessive exposure to AI‑generated images can dull visual discernment.

Job Market Impact : AIGC reshapes job roles, reducing some tasks while creating new workflow‑centric positions.

Overall, the Cyber project illustrates how AIGC can transform image production pipelines, but also raises important considerations around creativity, ethics, and workforce evolution.