Ask Yourself Three Questions First
- Do you want it to look good or work well? For posters and concept art where quality comes first → Midjourney; for background removal, local edits, and batch production where control matters → Stable Diffusion.
- Are you comfortable writing prompts in English? If you want to write prompts in Chinese and skip learning parameters → Dreamina; if you're happy tuning parameters in English → Midjourney / Stable Diffusion.
- What about your machine and budget? No setup, just pay monthly → Midjourney / Dreamina; have a GPU and want zero subscription fees → run Stable Diffusion locally.
How to Get Started with Each
① Midjourney — The Quality and Creativity Benchmark
Its visual aesthetics and quality lead the field, great for illustrations, posters, and concept art. It started on Discord but now has a web app; the core is writing prompts with /imagine.
Write a Prompt
Be as specific as possible: subject + environment + style + shot, then set the aspect ratio with --ar (e.g. --ar 16:9).
Generate a 4-grid, pick and upscale
It produces 4 images at once—pick one and Upscale, or use V to create variants.
Inpainting (optional)
Use the web app's selection inpainting to change a facial expression or background color without redoing the whole image.
Prompt you can copy directly:
② Dreamina — Chinese-friendly, Images + Video in One
From ByteDance, it generates images directly from Chinese prompts and even includes text-to-video—perfect for short-video covers and assets. Almost zero barrier for beginners.
Describe the scene in Chinese
Just write "what + style + purpose", e.g. a Xiaohongshu cover or an e-commerce hero image.
Choose aspect ratio and quantity
Use 9:16 portrait for phone covers, 16:9 landscape for illustrations, and generate several to choose from.
Convert to video (optional)
Turn a favorite image into a few seconds of motion video with one click—handy for short videos.
Prompt you can copy directly:
③ Stable Diffusion — Open-source, Controllable, Free Locally
Open-source and locally deployable, it offers extreme control through various models (checkpoints / LoRA) and node-based workflows—ideal for batch work, precise control, or those avoiding subscriptions. The trade-off is setup effort and GPU requirements.
Install a WebUI
Common options include Automatic1111 or ComfyUI; you'll need a decent GPU locally, or use cloud GPU if you'd rather not install.
Write prompt + negative prompt
English prompts are usually more stable; the negative prompt filters out issues like "blurry images, extra fingers, watermarks".
Swap models, tune parameters
Pick style models from sites like Civitai, adjust sampling steps, size, and seed to produce batches in a fixed style.
Prompt you can copy directly:
Quick Reference Table
| Dimension | Midjourney | Dreamina | Stable Diffusion |
|---|---|---|---|
| Strongest at | Quality / creative aesthetics | Chinese-friendly, image+video in one | Controllable / free locally |
| Ease of start | Low (just write prompts) | Very low (write in Chinese) | High (setup, learn parameters) |
| Cost | Subscription, see official site | Free tier available, see official site | Open-source free, but needs GPU/cloud cost |
| Best for | Designers, marketers, creators | Beginners, short-video creators | Developers, power users, batch production |
Four Pitfalls to Avoid
- Don't expect it to get text right: large text and slogans on posters are often wrong—always add text in design software afterward, don't let the AI force it.
- The more specific the prompt, the better: "a pretty cat" is worse than "an orange cat lounging on a windowsill, backlit, 35mm film look"—vague input only yields vague images.
- Mind copyright and commercial use: before commercial production, read each platform's terms and model licenses to avoid trouble.
- Don't pay right away: start with free tiers or Dreamina to test the waters; only consider a Midjourney subscription once you're sure the style fits.
Back to the homepage for more reviews, or browse AI art tools to find one that fits.
← Back to Reviews