為什麼我的 AI 繪圖很醜？

多半因為提示詞太籠統。你寫一隻貓，AI 只能給平均的貓。給的細節越多（主體、風格、光線、構圖），畫得越接近你腦中的畫面。

怎麼寫好 AI 繪圖提示詞？

套公式：主體（具體描述）＋動作情境＋風格＋光線氛圍＋鏡頭構圖。把這幾塊填滿，圖立刻不一樣。

加風格真的有差嗎？

差很多，是最影響成果的一招。同樣主體加油畫風、日系動漫、3D 渲染、電影感，出來完全不同，想要什麼感覺直接講。

AI 繪圖第一張就要滿意嗎？

不用。AI 繪圖是調出來的不是一次到位，看哪裡不對改提示詞那一塊，生圖、看、改、再生，幾輪下來就會抓到感覺。

AI 繪圖一直生出醜圖？問題不在工具，在你的「咒語」怎麼下

很多人用 AI 繪圖生出一堆四不像，就覺得 AI 繪圖不行。其實同一個工具，會下提示詞和不會的人，生出來的圖天差地遠。這篇教你怎麼把「咒語」下好。

I've seen too many people excitedly try out AI-generated art, only to produce a bunch of poorly proportioned and unrecognizable images, and then conclude that "AI art is nothing special." But with the same Midjourney tool, the difference between the images produced by experts and beginners is like night and day. The difference lies not in the tool, but in how you "cast your spell" (write your prompts). In this article, I'll share my experience with you.

Why are your images so ugly?

The most common reason: your prompts are too vague. If you simply type "a cat," the AI can only give you an "average cat" - one without style, emotion, or lighting. AI can't read minds, so the more details you provide, the closer the image will be to what you have in mind.

5 key factors to casting a good spell

1. Make your subject specific. Don't just write "a girl," write "a 20-year-old girl with long black hair, wearing a white dress, and smiling." The more specific your description, the more specific the image will be.

2. Add a "style". This is the most crucial factor in determining the outcome. With the same subject, adding "oil painting style," "Japanese anime style," "3D rendering," or "cinematic feel" can produce completely different results. If you want a specific feel, just say so.

3. Describe the "lighting and atmosphere". "Warm sunset light," "soft cloudy light," or "neon lights" - lighting and shadows determine the texture and mood of an image, so don't omit them.

4. Specify the "camera angle and composition". "Close-up," "wide-angle," "bird's eye view," or "shallow depth of field" - the AI understands these photography terms and can help you control the image.

5. Use "references" to communicate. If you want to capture the feel of a particular artist or image, describe its characteristics; many tools also allow you to upload reference images.

A practical formula

If you can't remember these factors, just follow this order: subject (who/what) + action/context + style + lighting and atmosphere + camera angle and composition. Fill in these elements, and your image will be transformed.

The most important mindset: iteration

It's normal if the first image doesn't meet your expectations. AI-generated art is "tuned" rather than "one-shot". Look at what's not right and adjust the corresponding part of the prompt - if the color is off, adjust the color; if the style is wrong, change the style; if the composition is odd, adjust the camera angle. Generate an image, review it, make adjustments, and generate again. After a few rounds, you'll get a feel for it.

Don't blame AI art for not being good enough. It's actually very powerful, but it's waiting for you to speak clearly. Choose a tool and try again using the methods above - you won't go back. For further reading: AI Art Generation Beginner's Guide, Midjourney Advanced Tips.

In a nutshell: Most ugly AI-generated images are not due to the tool being inferior, but rather the prompts being too vague - make your subject specific, add style, describe lighting and camera angles, and iterate several times. With the same tool, you can produce expert-level images.

Sources

Compiled from the editorial team's actual experience.