Table of Contents
Quick Answer
Create professional AI art by using structured prompts (subject + style + mood + camera + lighting + quality tags) in Midjourney or DALL-E, then refine with variations.
- Prompt structure > prompt length
- 3-5 generations per idea minimum
- Style consistency needs --sref or style references
What You'll Need
- Midjourney subscription ($10/mo minimum) or DALL-E via ChatGPT Plus
- Discord account (for Midjourney)
- A creative vision or mood board
- Pinterest or PureRef for references
- Photoshop or GIMP (optional, for final touches)
Step 1: Set Up Midjourney
Join Midjourney Discord → subscribe → use /imagine command in any bot channel or your DMs. Or use the web UI at midjourney.com/explore.
Step 2: Master the Prompt Formula
Structure: [subject], [style], [mood], [camera/medium], [lighting], [detail], --ar [aspect ratio] --v 7
Example: "A lone astronaut on a Martian dune at sunset, cinematic oil painting, melancholic, wide-angle lens, rim lighting, hyperdetailed dust particles --ar 16:9 --v 7"
Step 3: Iterate With Variations
After your 4-grid generates: click V1/V2/V3/V4 to vary your favorite. Click U1/U4 to upscale. Repeat 3-5 times.
Step 4: Use Style References for Consistency
For multiple images in same style: upload a reference image URL → add --sref [URL] to your prompt. All images share the aesthetic.
Step 5: Character Consistency
Use --cref [character image URL] to keep a character consistent across scenes. Critical for comics, children's books, branding.
Step 6: Refine in Photoshop (Optional)
Export PNG → open in Photoshop → use Generative Fill to edit specific regions → add text or composite with real photos.
Step 7: Manage Rights + Licensing
- Midjourney Pro plan: commercial use included
- DALL-E via ChatGPT: commercial use allowed
- Always check latest terms — they change
- Copyright office requires human creative direction for ©
Common Mistakes to Avoid
- Over-long prompts (Midjourney weighs first 40 words heaviest)
- Ignoring aspect ratio (--ar 1:1 default, use 16:9 for banners)
- Settling on first generation (always iterate)
- Publishing without disclosure in art communities
- Copying living artists' names in prompts (ethically + legally risky)
Top Tools
Tool
Use Case
Free Tier
Best For
Midjourney
Premium AI art
No ($10+/mo)
Artistic quality
DALL-E 3
Integrated with ChatGPT
Paid
Text-in-image
Stable Diffusion
Self-hosted
Yes
Full control
Leonardo.ai
Game assets
Yes
Game + UI
Krea
Real-time AI
Yes
Iteration speed
FAQs
Is AI art "real" art? Debated. Practically: galleries and publishers now accept it with disclosure. Ethically: use your own prompting voice, not copied names.
Can I copyright AI art? US Copyright Office (2023): AI-generated images alone can't be copyrighted. Human-authored elements can. Always add substantial human editing.
Midjourney vs DALL-E? Midjourney = artistic quality. DALL-E = text accuracy + integration with ChatGPT.
How many variations should I run? 3-5 minimum. Best results often come from generation 8-12.
Can I recreate a copyrighted character? No — infringement regardless of AI use.
What's the best aspect ratio? 1:1 Instagram, 16:9 banners, 2:3 prints, 9:16 phones.
Do I need Photoshop? Not required. Krea and Magnific handle most refinements now.
Conclusion
AI art is a new medium — like photography 180 years ago. The tools are cheap; the taste is still rare. Develop yours through thousands of iterations.
Try Assisters free →