AI tools aren't "type-and-it-figures-it-out". Each one has different strengths — pick the right one first:

• Product photos, people, scene shots → Midjourney v7 or Flux Pro
• IG quote cards / posters with text → Ideogram 3.0 (only model that renders text reliably)
• Customisation, full control, ComfyUI → Stable Diffusion 3.5
• Background music, product soundtracks, ballads → Suno v5.5
• Cinematic short clips → Sora 2 / Veo 3.1 / Kling 2.0
• Copywriting, resume editing, writing → ChatGPT / Claude / Gemini

Not sure which? See the /en/models comparison.

The fastest way for beginners — copy / paste, see results:

1. Clean product photo on white — Midjourney: learn the basic prompt structure (subject + style + lighting + composition).
2. Bilingual IG quote card — Ideogram 3.0: learn the "variable" concept ([zh_quote] / [en_quote] swap in your own content).
3. Café lo-fi 30s — Suno: learn the music prompt pattern (genre + tempo + mood + structure).

Once you've run those three, you grasp ~70% of prompt grammar. Head to /en/prompts and filter by tool or scenario for the next one.

Before you skim 200 prompts, internalise these 5 skeletons — everything else is variation:

• Natural language sentence: Flux / Suno prefer it. "a golden retriever puppy sitting on a wooden floor, soft window light"
• Comma-separated tags: Midjourney / SD prefer it. "golden retriever, wooden floor, window light, 35mm film, --ar 16:9"
• Structured template: Ideogram / poster work. "Subject: X. Style: Y. Layout: Z. Color: A. ar 1:1"
• Storytelling for video: Sora / Veo prefer it. "Camera slowly pushes in on..."
• Song-structure for music: Suno custom mode. "[Verse] ... [Chorus] ... [Bridge] ..."

Real scenario. You want a 30-second IG reel introducing a product. The full workflow:

1. Use Midjourney to generate 5 different product angles
2. Use Sora 2 to turn 3 of them into 5-second clips
3. Use Suno to generate a 30-second lo-fi soundtrack
4. Use ChatGPT to write the caption + hashtags
5. Composite in CapCut / Premiere

PromptCraft has a prompt template for every step. Check the "collections" hub for full multi-prompt workflows in one bundle.

• Too abstract: "a beautiful image" — the model can't infer that. Try "a minimalist Japanese tea cup, top-down view, soft natural light, beige background".
• Wrong tool: asking Midjourney to render CJK text → fails. Use Ideogram for text-in-image.
• Skipping the sample: every prompt page has a sample output. Read it before running so you don't waste credits.
• Not editing variables: copying a template without replacing [variable] tokens. Variables exist for you to plug in your own content.
• Quitting too early: the first roll rarely matches the sample. Tweak 1–2 keywords and rerun 1–2 times — usually converges.