---

個人感受

AI擅長的主題 + 好的prompt + 局部重繪 + 后處理 = 好的出圖效果

AI出圖效果的好壞強依賴于prompt（提示詞），直接根據某個寬泛的主題出圖的效果通常很差，部分公眾號等媒體在宣傳中往往夸大了AI在科研繪圖中的作用。

AI技術能夠生成具有氛圍感、真實感和藝術性的插圖，作為科普插圖是足夠的。

然而，目前AI生成的插圖在精確性、可控性和科學規范性方面仍存在一定局限，因此難以直接應用于嚴謹的科研論文插圖中。

一些比較滿意的出圖

【風化侵蝕】

【凍土退化】

一些效果差的出圖，缺乏邏輯

一、AI繪圖流程

1.1 Midjourney

（1）環境配置

環境配置流程

1. 開通會員賬號/租共享賬號
2. 安裝discord
3. 創建服務器
4. 添加Midjourney Bot至群組
5. 用/imagine命令開始繪圖

（2）生成prompt

方式1：用Midjourney的/describe命令

方式2：上傳圖片至chatgpt生成繪圖所需prompt

假如我想生成幾乎一模一樣的圖片，請你給我這副圖片的prompt
請詳細的描述一下這張圖片，生成prompt。以便我重新繪畫，（可以忽略文
字）

（3）完善prompt

Midjourney的prompt分為三部分：

• 圖片URL
• 文字
• 參數

圖片URL通過上傳圖片到當前服務器，復制鏈接獲得，添加圖片URL生成的圖片在風格上會更貼近參考圖。
文字prompt的獲得見前文。
一些常用的參數有，其余參數在本文繪圖中保持默認，見下圖

--ar: 改變縱橫比
--no：設置否定詞，排除要素

（4）開始繪圖

用空格作為分隔符，將上一步驟中三部分的prompt，進行拼接如：

運行命令后，會一次性生成四張圖

（5）后處理

Midjourney會一次性生成四張圖，四張圖的數字編號排列為

1	2
3	4

如果對生成的圖不滿意，可以

1. 調節prompt（這個最直接）
2. 點擊刷新，重新生成4張圖
   
3. 選擇V開頭的選項，選擇四張圖中你偏愛的風格，生成相近的四張圖
   
4. 局部重繪
   Midjourney有局部重繪工具vary (Region)
   
   開啟局部重繪窗口后，可選擇右下角的套索工具或矩形工具選擇要進行局部重繪的區域，基于prompt進行重新繪制。
   
   可以看到局部重繪后的圖片，僅在被選中的修改區域產生了變化，非選中區域則基本不變
   
5. 修改結果圖比例
   點擊custom zoom工具
   
   在原有prompt的基礎上，修改參數項，為

# 命令范式
# --ar 目標圖片比例 --zoom 1# 將圖片輸出尺寸修改為1:1，整體縮放不變
--ar 1:1 --zoom 1

修改尺寸后的繪圖結果為

單擊圖片后，右鍵下載圖片到本地

1.2 ChatGPT

ChatGPT的生圖功能基于OpenAI的DALL·E模型，普通用戶使用該功能限制為2張/24h，升級Plus賬號后限制有放寬。

不合理的出圖結果

本人使用體驗ChatGPT生圖功能遇到的錯誤有
（1）文字、箭頭的錯誤

（2）無關的裝飾要素過多

（3）對地下部分的描述脫離現實
地下生長的植物

輪船航行在地下

奇怪的地下結構

植物莖干部分和地下根系的錯位

2D profile showing a mixture of vegetation with different root depths. Grass with shallow roots about 10 cm deep, shrub with medium roots about 50 cm deep, and tree with deep roots about 3 meters deep. Multiple soil layers visible beneath the vegetation.

解決方案

1. 降低文字生成的可能，凸顯主題
   ChatGPT的繪圖不支持排除項的指定，如指定出圖不包含某個元素。因此直接設置出圖結果不包含文字、箭頭等，效果并不好。但可以通過輸入肯定的prompt來凸顯出主體，prompt如下

[global option] Focus on specific, visually representable elements. Describe actions and scenarios rather than abstract concepts. Avoid ambiguous language that could be interpreted as including text.

2. 出圖后用局部重繪功能，remove去除不想要的元素
   

---

二、主題繪圖結果展示

地球內部圈層

An artistic cross-sectional diagram of the Earth showing its internal layers, including the crust, mantle, outer core, and inner core. Each layer is vividly colored, with distinct textures and gradients to represent the density and composition changes. The background is white, and the Earth is partially transparent to reveal the layers within. The inner core glows with a bright yellowish light, representing its heat and solid state.

【ChatGPT】

【Midjourney】

史前時期地貌演化模式

A detailed artistic diagram illustrating the evolution of prehistoric landscapes. The image is a spiraling timeline showing changes in terrain over geological periods, including mountains, rivers, forests, and deserts. Each segment of the spiral represents a distinct geological era, with vivid details like volcanic eruptions, glacial formations, and vegetation development. The background is white to highlight the colorful and intricate layers of terrain evolution.

【ChatGPT】

【Midjourney】

不同時期化石演化

A detailed artistic illustration showcasing fossil evolution across different geological periods. The image consists of a series of layered blocks, each representing a distinct time period, with stratified earth layers and corresponding surface ecosystems. Fossilized remains of plants and animals are depicted in each layer, showing gradual changes over time, such as dinosaurs, mammoths, and early human activity. The background is white, emphasizing the colorful and detailed transitions of geological and biological history.

【ChatGPT】

【Midjourney】

板塊運動

Continent movement
A detailed cross-sectional diagram of Earth’s lithosphere showcasing plate tectonics. The image includes mountain formations, subduction zones, mid-ocean ridges, and volcanic activity. The layers of the Earth’s crust and mantle are clearly depicted with distinct textures and colors. The background is white, emphasizing the dynamic processes of plate movement, such as divergence, convergence, and magma rising at the mid-ocean ridge.

【ChatGPT】

【Midjourney】

地質活動

2D profile, geological activity at a mid-ocean ridge. The diagram depicts two tectonic plates moving apart due to magma rising from the mantle, forming new oceanic crust. The Earth’s layers are represented with distinct textures and colors, showing the crust, mantle, and magma. The background is white, emphasizing the dynamic process of seafloor spreading.

【ChatGPT】

【Midjourney】

地層褶皺

Stratigraphic folding in the Earth’s crust. The illustration features layered sedimentary rocks bent into an anticline and syncline structure. The layers are shown in different colors to represent their composition and depth. The surface is green, symbolizing vegetation, and the background is white to emphasize the geological deformation.

【ChatGPT】

【Midjourney】

地震

The aftermath of an earthquake in an urban setting. Collapsed buildings, tilted structures, cracked streets, and a derailed tram. Smoke and fire rise from destroyed buildings in the background, while people are depicted in chaos and rescue efforts. The destructive power of earthquakes with intricate details and a white background.

【ChatGPT】

【Midjourney】

海嘯

a tsunami caused by an underwater earthquake. The image shows the ocean floor with a fault line, the displacement of water due to seismic activity, and the resulting waves propagating toward the coastline. Palm trees and small huts on the shore highlight the vulnerability of coastal areas. The background is white, emphasizing the dynamic process of wave formation and energy transfer.

【ChatGPT】

【Midjourney】

雪崩

A dramatic scene of a massive avalanche descending from a towering, snow-covered mountain. The avalanche rushes down the slope with immense force, its dense snow cloud and debris cascading toward the valley below. At the foot of the mountain, a pine forest surrounds a few houses, with people running in panic to escape the oncoming disaster. Animals, including deer and birds, are seen fleeing the area. The snow is already beginning to engulf parts of the landscape, creating a sense of chaos and urgency. The lighting is natural but slightly overcast, with a cold, white-dominated palette emphasizing the snow and tension in the atmosphere. Created using: cinematic composition, dynamic motion effects, realistic textures, vivid environmental details, high-definition quality, dramatic lighting, and an intense, natural disaster theme.

【ChatGPT】

【Midjourney】

火山噴發

A simplified illustration of a volcanic eruption showing a cross-sectional view of a volcano. The diagram features a cone-shaped volcanic mountain with lava flowing down its slopes and thick smoke and ash rising into the air. Surrounding the volcano are small patches of greenery and a water body at the base. The background is white, emphasizing the eruption process and the volcano’s structure.

【ChatGPT】

【Midjourney】

巖漿流動

A detailed cross-sectional illustration of magma flow, showing molten lava moving through volcanic channels and erupting on the surface. The diagram highlights the underground magma chamber feeding the lava flow, with bright orange and red tones representing heat and molten rock. The surface features fiery explosions and glowing lava spreading across rugged terrain. The background is white, emphasizing the dynamics of magma movement.

【ChatGPT】

【Midjourney】

冰川地貌

Glacial landforms in a mountainous region. The image features U-shaped valleys, cirques, tarns, and rivers flowing through the valleys. The terrain is rugged, with steep mountain peaks and lush green vegetation on the slopes. Small lakes are scattered in the valleys, connected by streams. The background is neutral, emphasizing the geological features shaped by glacial activity.

【ChatGPT】

【Midjourney】

河流地貌-細小河流匯聚

A detailed isometric illustration showing a river system in a mountainous region. The image features snow-capped peaks, dense vegetation, and multiple streams converging into a main river channel. The terrain is rugged with steep slopes and carved valleys. The river is depicted in blue, flowing dynamically through the landscape. The background is neutral, emphasizing the natural river system.

【ChatGPT】

【Midjourney】

河流地貌-河流穿過樹林

A detailed isometric illustration of a meandering river in a forested landscape. The river is shown in blue, curving gently between lush green forests on both sides. The terrain features a cross-section of soil layers, emphasizing the riverbank’s structure. The trees are dense, creating a natural, serene environment. The background is neutral, focusing on the river’s flow and surrounding vegetation.

【ChatGPT】

【Midjourney】

河流地貌-上游至下游

a river’s journey from upstream to downstream. The image features snow-capped mountains at the source, a dam controlling water flow, and the river winding through various landscapes. Surrounding elements include dense forests, agricultural fields, orchards, bridges, and a city at the downstream end. The terrain showcases cross-sections of soil and rock layers, emphasizing the connection between natural and human-made features. The background is white, highlighting the progression of the river through the environment.

【ChatGPT】

【Midjourney】

河流地貌-河間地塊

A simplified cross-sectional diagram illustrating fluvial landforms. The image shows valleys carved by river erosion, with U-shaped and V-shaped channels on the surface. Thin blue rivers flow through the valleys, highlighting the process of erosion and sediment transport. The terrain is composed of sandy or soil-like material, and the background is white to emphasize the geological features.

【ChatGPT】

【Midjourney】

喀斯特地貌

A detailed cross-sectional illustration of karst landforms, showcasing a landscape shaped by water erosion and dissolution of limestone. The diagram features sinkholes, underground rivers, caves, and cracks in the rock layers. Water flows through the system, creating interconnected channels and reservoirs. The surface includes grasslands and small streams. The background is white, emphasizing the internal and external features of the karst system.

【ChatGPT】

【Midjourney】

風化侵蝕

Weathering and Erosion
A detailed isometric illustration depicting a landscape shaped by weathering and erosion. The image features eroded rock formations, layered sedimentary structures, and a desert-like terrain. The terrain includes canyons, mesas, and a small basin filled with water. Soil and rock layers are exposed, highlighting the effects of natural forces over time. The background is white, emphasizing the geological processes that shaped the land.

【ChatGPT】

【Midjourney】

水循環

A detailed isometric illustration of the terrestrial hydrological cycle, featuring precipitation, surface water evaporation, groundwater flow, vegetation transpiration, and solar radiation. The diagram includes mountains, rivers, forests, and clouds. The sun is depicted as the primary energy source driving evaporation and transpiration. The background is white, emphasizing the interconnected processes of the water cycle.

【ChatGPT】

【Midjourney】

土壤剖面1

A detailed illustration of a soil profile featuring three distinct layers. The top layer shows plants with green leaves and roots extending into the soil. The soil layers are depicted with different textures and colors, ranging from dark, organic-rich topsoil to lighter subsoil and coarse, rocky material at the bottom. The roots penetrate through all three layers, connecting the vegetation to the soil. The background is white, emphasizing the soil structure and plant interaction.

【ChatGPT】

【Midjourney】

土壤剖面2

A detailed isometric illustration of soil layers from top to bottom, including humus, topsoil, subsoil, weathered rock fragments, and bedrock. The surface features a tree with deep roots extending into the subsoil and grass with shallow roots confined to the humus layer. Each soil layer is distinctly colored and textured to show its composition. The background is white, emphasizing the stratification and root interactions.

【ChatGPT】

【Midjourney】

土壤質地方塊

An educational illustration of three vertical rectangular prisms placed side by side, representing different soil textures. The left prism shows salinized soil with a pale crust and dry, dead grass on top. The middle prism depicts fertile brown-yellow soil with green herbaceous vegetation on the surface. The right prism represents arid, cracked soil with visible fractures and no vegetation. The design is minimalistic, focusing on the textures and colors of the soil, with a clean white background.

孔隙含水介質

Porous Aquifer Medium
A simplified 3D illustration of a porous aquifer medium. The diagram shows a cube filled with interconnected pores and solid grains, representing the spaces where water can flow and be stored. The background is neutral blue to emphasize the porous structure and the contrast between the solid material and the voids. Each pore is highlighted to show water retention and movement potential.

【ChatGPT】

【Midjourney】

油氣開采

A detailed cross-sectional illustration of an oil and gas extraction site. The surface includes a drilling rig, storage tanks, and infrastructure set on a desert landscape. Below the surface, multiple geological layers are shown, with a drilling well extending through the layers to reach the oil and gas reservoir. The reservoir is depicted as a black layer trapped between impermeable rock layers. The background is white, emphasizing the subsurface and drilling process.

【ChatGPT】

【Midjourney】

衛星遙感

Satellite remote sensing,real photo

【ChatGPT】

【Midjourney】

水庫

an illustration showcasing watershed water resource management. A dam intercepts the river, storing water in an upstream reservoir. The reservoir’s bottom consists of bedrock, and water is regulated through the dam’s gates before being discharged into the downstream plain area.

【ChatGPT】

【Midjourney】

地面沉降

land subsidence caused by excessive groundwater extraction. The image features a coastal area with tilted and sinking buildings, cracked ground, and a lowered surface layer. Arrows indicate the upward flow of water from underground, and the subsurface layers show depleted aquifers. The background includes a mix of land and water, emphasizing the environmental impacts of over-extraction on urban and rural landscapes.

【ChatGPT】

【Midjourney】

---

三、出圖效果對比

（1）出圖效果對比
Midjourney的效果整體比ChatGPT更好，速度更快，更加方便二次修改
（2）成本對比
官方售價均在人民幣100元左右，二次市場的共享賬號方面，Midjourney略便宜。
【ChatGPT會員】

【Midjourney會員】

（3）風格差異
ChatGPT的繪圖風格是綺麗絢爛的夢，通常具有較高的飽和度，色彩鮮明、對比度較強

MidJourney的繪圖風格則相對柔和，飽和度較低，色調更為平衡一些

---

參考鏈接
1.[60張高清地質用圖] https://mp.weixin.qq.com/s/spozxpFLvkstA7wZOZKAsA