Illustrated portrait of Shen Jiaying, an AI tutor

AI tutor

Shen Jiaying AI

Image generation run as an experiment: fixed seeds, one change, a written log.

Teaches in English · 简体中文

Ask ShenHow Shen explains

How Shen explains

I keep a four-line log open beside every session, and I ask you to fill it in before you close the window. In Image Generation, in Practice I separate two modes clearly: random seeds for exploring what a prompt can do, a locked seed for testing one change. I refuse to judge a prompt edit made on a random seed, because nothing can be learned from it. In Editing an Image I finish the chosen frame by hand in Photopea or GIMP. In AI for a Small Business I cost a batch of catalogue images before we generate any of them.

Suits you if you generate product images for a shop and cannot tell whether your last prompt change helped at all.

A sample exchange

  1. 我改了提示词里的一个词，图片好像好看了一点。这能说明这个改动有效吗？

  2. Shen Jiaying AI

    如果种子（seed）是随机的，就说明不了。种子决定生成开始时的那片随机噪声：同一个种子、同样的设置，出来的就是同一张图。你同时换了种子和提示词（prompt），就无法判断差别来自哪一个。探索时用随机种子，测试时锁定种子，只改一处再对比。另外，种子只在同一套配置下才能复现，模型、采样器、分辨率任何一项变了，结果都会不同。每次收工前，把最好的两三个种子记进项目日志，各写一句说明。1

    1. Image Generation, in Practice · Lesson 3.6 · Fix the seed or learn nothing

A sample, grounded in that lesson. Your own question gets its own answer.

Courses Shen answers for

Ask ShenSee all AI tutors