AI Image Prompts That Actually Work — People & Animals (GPT, Gemini, Midjourney)
Why do some AI images look great and yours miss the mark? Mostly the prompt. Learn the simple five-part structure behind good prompts, copy-ready templates for people and animals, what differs between GPT, Gemini and Midjourney — then how to cut out the subject and clean up the result with free browser tools.
How AI image generation actually works (in one minute)
When you type a prompt and hit enter, what happens inside? Most image models today are "diffusion models": they start from pure random noise and, guided by your text, gradually erase the noise and shape a picture over many steps. ChatGPT and Gemini are language models that can also draw; Midjourney is a specialist whose whole job is generating images. Your prompt is the steering wheel — not a magic spell. The more specific and concrete your description, the closer the model lands.
- Diffusion = start from noise, gradually shape it into a picture using your text as a guide
- GPT / Gemini: general-purpose language models that also generate images — great with full sentences
- Midjourney: a dedicated image model — great at style, needs its own short-phrase format
- Rule of thumb: a vague prompt gives a vague image; a specific prompt gives a useful one
The five-part structure of a good prompt
Every strong prompt can be broken into five parts: (1) the subject, (2) its details and features, (3) the action or pose, (4) the setting and background, (5) the style and quality. In Chinese that looks like: 「主体 + 细节特征 + 动作姿势 + 环境背景 + 风格画质」. You do not need all five every time, but the more of them you fill in, the more control you have.
- Subject: a woman / a cat / a city — who or what is in the picture
- Details: age, hair, clothes, fur, color, expression — what makes it specific
- Action: sitting, running, waving, looking at the camera
- Setting: on a beach, in a studio, snowy forest, plain background
- Style & quality: photorealistic, anime, watercolor, 8k, cinematic lighting
How to write prompts for people
For people, the model needs enough detail to know exactly who to draw. Go down the list: gender and age, face shape and facial features, hairstyle and hair color, clothes, pose, expression, lighting, and camera/angle. Replace "a girl" with a concrete description and the difference is night and day. For example: 「30 岁的中国女性，自然淡妆，黑色长直发，白色衬衫，微笑着看向镜头，柔和自然光，浅灰纯色背景，半身肖像照，真实摄影，细节丰富」.
- Face: mention age, face shape, hairstyle, hair color, glasses if any
- Expression & pose: smiling, serious, looking away, hands in pockets
- Clothing: specific beats vague — "white shirt" beats "clothes"
- Quality words: photorealistic, high detail, 8k, cinematic lighting
- Watch out: hands and eyes are where models stumble — fewer fancy hand poses, more front-facing
- For a consistent character across images, keep the character part of the prompt fixed and change only the action
How to write prompts for animals
Animals follow the same five-part structure, but the "details" slot is about species and breed, fur color and pattern, and the distinctive features that make the animal recognizable. A pet photo wants "cute, warm home light, shallow depth of field"; a wildlife shot wants "natural light, telephoto, sharp". Keep the species name in the prompt — "a ginger British Shorthair" beats "a cat" every time.
- Name the species AND the breed: "ginger British Shorthair" not "cat"
- Fur: color, pattern (striped, spotted), length (short, long, fluffy)
- Action: sleeping, running, sitting, looking up, begging
- Style switch: photorealistic pet photo / anime style / watercolor illustration
- Example: "a golden retriever puppy running on a sunny lawn, natural light, sharp, high-detail photography"
GPT, Gemini and Midjourney — how the writing style differs
The same idea needs three slightly different ways of writing. In ChatGPT (DALL·E) and Gemini, write a full natural-language paragraph and you can even have a back-and-forth — "make the background plain", "change the shirt to red" — refining in conversation. Midjourney uses /imagine followed by short comma-separated phrases plus parameters: --ar 16:9 for aspect ratio, --v for model version, --style for a stylized look. Both styles work; just match the tool.
- ChatGPT (DALL·E): full sentences, very natural-language friendly, supports Chinese well
- Gemini: full sentences, and you can iterate in the same chat ("make it brighter")
- Midjourney: /imagine short phrases + parameters (--ar 16:9, --v 6, --q 2)
- Same prompt idea — Midjourney form: "portrait of a smiling woman, white shirt, soft light, plain background, photorealistic --ar 3:4"
Copy-ready templates for people and animals
Here are ready-to-use templates. Copy one, swap the details in the first two slots, and you are most of the way there. Keep the last slot (style + quality) when you like a look, and delete it when you want to try a different style.
- Portrait: "a 30-year-old Chinese woman, light makeup, long straight black hair, white shirt, soft smile, looking at camera, soft natural light, light gray plain background, half-body portrait, photorealistic, high detail"
- Anime girl: "anime style girl, twin tails, pink hair, purple eyes, sailor uniform, side profile, sunlight, cherry blossom background, high-detail illustration"
- Headshot: "professional headshot, man in his 40s, short hair, navy suit, confident smile, plain background, studio lighting"
- Cat: "a ginger British Shorthair, round face, big eyes, sitting upright, warm indoor light, photorealistic pet photography"
- Dog: "a golden retriever puppy running on a sunny lawn, natural light, sharp, high-detail photography"
- Wildlife: "a Siberian tiger walking through snow, sharp eyes, natural light, wildlife photography"
Why AI images usually still need a little cleanup
Even a great AI image rarely comes out perfect for the use you have in mind. The background is messy or busy, the edge of the subject has a thin white fringe, there is a text watermark or logo in a corner, or the resolution does not match where you want to post it. These are exactly the jobs a quick browser tool does in seconds — and doing them yourself is far easier than trying to get the AI to fix them.
Clean up with SuperPixMia — cutout, new background, watermark
All the tools below run 100% in your browser — your AI image never gets uploaded. To swap the background: open the Remove Background tool, it cuts the person or animal out automatically and gives you a transparent PNG, then paste it onto any new background. To cut precisely by hand: open Studio and use Cut Out (trace a loop) or the magic-wand style Remove tool (click to select similar pixels). To fix a white fringe along the edge, click the fringe with Remove and it swallows it. To remove a watermark, logo or stray text, use the Remove Watermark tool (paint over it) or the Heal / Clone brushes in Studio. When you are done, resize, compress or convert to the format you need — every one of these tools also handles a whole batch at once.
- New background: Remove Background → transparent PNG → place on any new background
- Precise cutout: Studio → Cut Out (trace) or Remove (magic wand click)
- White fringe on the edge: click it with the Remove tool — it selects and clears it
- Watermark / text: Remove Watermark (paint over) or Studio Heal & Clone brushes
- Size & format: Resize, Compress, Convert — all also work in batch
Frequently Asked Questions
Which AI image tool is best for a beginner?
If you already use ChatGPT or Gemini, start there — natural language, easy to iterate in the same chat. If you want maximum style control, Midjourney is worth the learning curve. Whichever you pick, the prompt structure in this guide works for all of them.
Should I write prompts in Chinese or English?
ChatGPT and Gemini handle Chinese well, so write whatever you are comfortable with. Midjourney understands Chinese but tends to perform best with English short phrases. If results feel off, translating the last part of your prompt to English usually helps.
Why do AI faces and hands still look weird?
Faces and hands are the hardest parts for image models — there is no prompt that fully guarantees them. Reduce fancy hand poses, favor front-facing faces, and if a detail is wrong, regenerate or fix it later in a tool (paint over it) rather than fighting the model.
After cutting out the subject, how do I put it on a new background?
Run Remove Background to get a transparent PNG, open it in any editor (or your messaging app), and paste it over your new background. For a matching soft edge, cut out with Studio instead of the auto tool for fine control.
The generated image has a white fringe around the subject. How do I clean it?
Open the image in Studio, switch to the Remove tool, and click on the white fringe — the magic wand selects the thin white ring around the subject and clears it. Raise the tolerance if it misses thin areas.
More from the blog
Why WeChat Images Look Blurry — 5 Causes & Exact Fixes
WeChat makes your photos look blurry? It is not your phone. Auto-compression, the 25MB limit, and the "send original" checkbox are the real causes. Here are the 5 reasons and the exact fix for each.
How Does AI See Images? Pixels, Tensors & Neural Networks Explained
When you see a cat, an AI model sees a grid of numbers. A beginner-friendly walkthrough of how images become tensors, why convolutions look at small windows, and how stacked layers turn edges into full objects.
How to Make AI Animated GIFs (Stickers & Memes): AI Frames → Batch Cleanup → Assemble
The frame-by-frame method: ask an AI to draw 4–6 frames of the same character in different poses, batch-clean every frame (cutout, watermark, uniform size) with free browser tools, then combine them into an animated GIF. No video model required.
