In one sentence: the gap between pros and everyone else isn’t the first prompt — it’s the ability to write the second one based on the first result. Our cards give you a high-quality starting point; this guide shows you how to make them yours.
New here? Read the beginner tutorial first (copy → paste → generate). This guide is for people who have already generated a few images.
1️⃣ Turn card prompts into “editable templates”
Every prompt on this site is “skeleton + variables”. Take the avatar card:
A vibrant anime-style portrait of a young girl with flowing silver hair...
(silver hair / pastel color palette are the swappable variables)- Swap the subject: girl → boy / cat / robot — any gender, any species
- Swap colors: silver hair → red hair — color changes apply globally
- Swap style words: anime → watercolor / 3d render / pixel art
- The pattern: split “subject + environment + light + composition + style + quality” into six parts and touch only one or two at a time — results stay predictable
2️⃣ The five-iteration method (the core technique)
| Round | What to do | Example |
|---|---|---|
| 1st | Paste the card prompt as-is, see what you get | Verbatim |
| 2nd | Change ONE variable (usually subject or color) | girl → boy |
| 3rd | Point-fix what you dislike | add “detailed eyes” (blurry eyes → call out the eyes) |
| 4th | Adjust the mood (light / color / background) | add “golden hour, sunset glow” |
| 5th | Lock in the version you like — save it as your own style template | Keep for reuse |
Key principle: change one thing per round, otherwise you’ll never know which edit did what.
3️⃣ Quick troubleshooting table
| Problem | Cause | Fix |
|---|---|---|
| Malformed hands / broken faces | Not described — the model improvises | Add “perfect hands, symmetrical face” or explicitly “no deformities” |
| Wrong colors | Color words too abstract | Use concrete color language (“vibrant red dress”) instead of “nice colors” |
| Composition too empty / crowded | No composition words | Add “centered composition / close-up / wide shot” |
| Too “plastic” | Missing material/texture description | Add “photorealistic, film grain, 85mm lens” for realism |
| Multiple subjects blending | Unclear subject description | Set the hierarchy: “a single fox in the center” |
| Garbled text | AI renders text poorly | Spell out the exact copy + “text reads: XXX” (still needs touch-up after) |
4️⃣ Three word types that make images “tell a story”
- Mood words: serene / dynamic / nostalgic — set the emotional tone
- Camera words: cinematic lighting / shallow depth of field / Dutch angle
- Detail words: volumetric light / reflective surface / intricate details
Add 1–2 word types at a time, watch what changes — stacking everything at once makes images messy.
5️⃣ Special tricks for conversational generation
For conversational tools like ChatGPT (free tier works, no setup) — different from Midjourney-style commands:
- Natural language works better: “translating” the prompt into one fluent sentence is fine — the model understands semantics
- Keep editing in the conversation: “make the hair red”, “swap the background to a city night” — no need to rewrite the whole prompt
- Multiple options at once: ask for “4 variations”, compare quickly, then refine
- Avoid negative phrasing: conversational models handle “don’t do X” unstably — prefer positive descriptions (e.g. “no busy background” → “clean minimal background”)
Further reading
- The full beginner flow: the beginner tutorial
- Jump straight in: the Image Prompt Gallery