How Front-End Developers Can Ride the Waves in AIGC

AliMei’s Guide

The author has been working on AIGC-related projects since July and has gathered some insights and experiences to share with everyone. The improvement in the quality of generated images comes from the rapid development of large models and open-source plugins in the AIGC field, as well as a deeper understanding of the generation process and parameters.

How Front-End Developers Can Ride the Waves in AIGC

What You Think Is Riding the Waves Is Actually “Riding the Waves”

It’s not about enjoying beautiful images on sunny ☀️ beaches 🏖️, but shivering in a stormy sea.

Introduction

Since starting AIGC-related projects in July, it has been 4 months. While organizing the generated documents, it is evident that the quality of generated images has significantly improved, leading to many insights. The enhancement in quality comes from the rapid development of large models and open-source plugins in the AIGC field, as well as a deeper understanding of the generation process and parameters. Therefore, I would like to summarize the project content and generation experiences from the past few months.

Original Image	Image Generated in July	Image Generated in October

Cyber Project Overview

Cyber is an internal project initiated by a technology platform, focusing on the engineering of AIGC, including outputting generation process as a service, model training, deployment, testing, and a series of solutions. Currently, the technology is relatively mature in AI modeling direction and intelligent background generation.

AI Models

Application Scenarios:

AI model generation can be both B2B and B2C. B2B can help merchants reduce commercial shooting costs and improve shooting efficiency. As long as merchants provide several flat images of clothing, or clothing on mannequins, or real people wearing clothing, different ethnic models and product images can be generated as needed. B2C also has many applications, such as Miaoya Camera, AI Fitting Room, etc.

Types of Technical Solutions:

Type

SD Lora

SD Impainting

Midjourney Base Image

VITON/Tryon GAN

Tryon Diffusion

Applicable Scope and Overview

Flat clothing

Requires multiple training inputs; output cannot guarantee complete similarity

Long user wait time

Mannequin/Real person wearing clothing

Can guarantee complete consistency in clothing details/patterns

Simple clothing in color+style

MJ generated model effects are relatively natural, with fewer bad cases. However, it cannot guarantee complete consistency in clothing details.

Previous generation furnace

Open-source

Can achieve multiple angles and forms

Next generation, based on diffusion models, utilizing two UNet diffusion models

Not yet open-sourced, developed by Google

Research data indicates better performance than VITON/Tryon GAN, etc.

Effect Overview:

Merchant Original Image

Cyber

Generated Image

Intelligent Background

Application Scenarios:

Intelligent backgrounds are generally aimed at still life, beautifying backgrounds for electronics, home goods, accessories, cosmetics, etc. By matching product categories, corresponding AI backgrounds are recommended.

This capability is one that merchants are relatively accepting of and frequently use, generally involving automatic/manual background removal + canvas positioning + SD impainting + image editing.

Effect Overview:

Merchant Original Image

(White Background + Cowhide)

Cyber

Generated Image

Current Experience

Industry Status

(The Industry Chain of AIGC, from AI-Generated Content (AIGC): A Survey)

The upstream of the AI industry includes data labeling, data supply, open-source algorithms, computer hardware, etc.; the midstream includes cloud computing, performance enhancement, and real-time interaction fields, such as large tech companies like Google and OpenAI; the downstream includes content generation platforms, content distribution platforms, content detection platforms, etc.

While the public is either excited or resistant, industry giants have already completed relevant layouts. Regardless of the progress of user acceptance, the chip and cloud service industries have already made substantial profits from AIGC.

User Acceptance

After AIGC became popular, not everyone may have used AIGC, but most people might have seen the results generated by AI. Some have created personal IPs through AIGC to become bloggers, while others have become bloggers by quickly identifying AI-generated content. So, can users in the e-commerce field accept AI-generated images? Can it positively influence users’ purchasing decisions? This question may still require time and data to answer. According to existing data, beautified images have indeed increased click-through rates to a certain extent. We also look forward to more data sharing from the AIGC in the domestic and international e-commerce industries.

Positive Attention on Social Media

Uploading AIGC-generated images and AIGC generation tutorials on social media has garnered considerable attention for many bloggers.

There are now many AI anchors and AI IPs that are beautiful, won’t fail, and can work 24 hours a day. Who wouldn’t love that!

Negative Feedback on Social Media

There are also some negative reviews, mainly concerning copyright issues arising from AIGC images or discrepancies in the e-commerce field.

Soul-Searching Question: What Can Front-End Developers Do in the AIGC Field?

Responsible for Generation Effects

Jobs arising from AIGC typically involve front-end developers being responsible for preliminary technical research, with designers expanding styles.

(1) Explore AI image generation processes to output stable generation workflows for different types of business scenarios.

(AI Model Generation Process Diagram)

(2) Explore multi-domain AI generation, such as pets, anime, video generation, etc.

Ancient Style	Anime	Animals	GIF	Still Life	Scenery	Logo	Real Person	Wallpaper

Open Components (Generation Interaction/Image Processing/Layer Merging/Canvas Dragging/Manual Background Removal, etc.), Engineering Products

AI-specific components, such as background removal, 3D open pose editor, image processing editor, canvas dragging for merging, etc.

AIGC-related components will be organized and published in the “Cyber Frontend Component Access Document” to explore smoother user interactions in the AIGC field.

Some Thoughts

Creative Exhaustion Caused by Copyright Issues

As a rapidly growing emerging field, the usage scenarios and generated works are limitless. However, it is undeniable that AI-generated large model training data relies on existing data accumulation. If everyone takes the shortcut of AIGC generation, who will patiently spend a lot of time and effort on innovation to create new knowledge wealth belonging to humanity? Of course, this issue has already attracted a lot of attention, and many copyright-related restrictions have emerged, leading to increasing regulation in the industry.

Aesthetic Fatigue

Perfection leads to imperfection, and designers in the project team have similar feelings. After seeing many AIGC images, there is a certain degree of aesthetic fatigue, making it hard to distinguish beauty from ugliness. This reminds me that the real world is attractive and creative precisely because of its diversity.

Do Job Opportunities Decrease or Increase?

The emergence of AIGC has made many industries feel threatened, leading to reduced positions and increased barriers to entry. However, many designers use AI as a great tool for enhancing creativity, producing refreshing works. I have seen many stunning images/videos generated using Photoshop + SD + Midjourney + DALL-E combinations on C station. In the world of AIGC, people are no longer just seeking images but are looking for workflows to produce images.

Welcome to Join the 【Aliyun Developer WeChat Group】 Reader Group

This is a communication space specifically for readers of the “Aliyun Developer” public account

💡 Here you can discuss technology and practices, and we will also regularly release group benefits and activities~

Welcome to add WeChat: argentinaliu to join the group chat

Application Scenarios:

Types of Technical Solutions:

Effect Overview:

Application Scenarios:

Effect Overview:

Responsible for Generation Effects

Open Components (Generation Interaction/Image Processing/Layer Merging/Canvas Dragging/Manual Background Removal, etc.), Engineering Products

Leave a Comment Cancel reply