GPT-Image-2 Case · Transformer Encoder–Decoder Architecture

A reference case of GPT-Image-2 featured in the "Research Paper Diagrams" showcase, suitable for retrieving generative effects in the "Transformer Encoder–Decoder Architecture" domain.

Author: AI Plus Lab

✦Results

◌Case Background

来自 GPT-Image2-Skill README 的精选展示条目，适合作为“研究论文图示 / Transformer 编码器–解码器架构”方向的站内参考案例。

⌘Prompt Content

Horizontal 16:9 academic concept diagram illustrating the Transformer Encoder-Decoder Architecture, NeurIPS final draft style. Vertically stacked on the left and right columns, separated by a dashed line in the middle.

Left column title: "ENCODER (×N)". Modules in order from bottom to top: "Input tokens" → "Input Embedding" → "+ Positional Encoding" → Dotted box "Encoder layer" containing "Multi-Head Self-Attention", "Add & Norm", "Feed-Forward", "Add & Norm", with slender, bent residual arrows around each sub-layer.

Right column title: "DECODER (×N)". Modules in order from bottom to top: "Output tokens (shifted right)" → "Output Embedding" → "+ Positional Encoding" → Dotted box "Decoder layer" containing "Masked Multi-Head Self-Attention", "Add & Norm", "Multi-Head Cross-Attention" (with an arrow from the top of the encoder labeled "keys, values" pointing here), "Add & Norm", "Feed-Forward", "Add & Norm". Above the decoder: "Linear", "Softmax", "Output probabilities".

Title: "Transformer: encoder–decoder with multi-head attention". Subtitle: "Vaswani et al., 2017".

✎Outcome Notes

This case has been migrated from the GPT-Image2-Skill README to AIPlusLab `/prompts`, facilitating direct search, browsing, and reuse within the site.