Từ thư viện prompt đến workbench: ảnh GPT Image đầu tiên trong 30 phút

locale: en title: ‘From Prompt Library to Workbench: Ship Your First GPT Image in 30 Minutes’ description: ‘Beginner tutorial: copy a scene prompt on GPTImage Prompts, generate in the gptimage App workbench, run one single-dimension edit, and export.’ date: 2026-09-11 author: ‘GPTImage Prompts’ level: beginner tags: [‘Beginner’, ‘Prompts’, ‘Workbench’, ‘Tutorial’]

This tutorial is for your first GPT Image 2 / gptimage 2 run: no fluff—just copy prompt → generate → one edit → export. After this, you can use the prompt library and deeper tutorials on your own.

Independent guide site — not OpenAI official. Workbench links go to the gptimage App.

What you will finish

About 20–30 minutes.

Step 1: Open the library and pick a scene

1. Go to the prompt library
2. Choose a familiar scene (start with ecommerce or social)
3. Click Copy prompt

Optional: Prompt hub guide.

Step 2: Open the workbench and paste

1. Click Start Creating Free / workbench CTA (opens the App)
2. Confirm a GPT Image-capable model (labels vary by product UI)
3. Paste the prompt
4. For posters, heavy text, or series intent, enable Thinking

Need help choosing Thinking? See Thinking mode tutorial.

Step 3: Generate and edit one dimension

Do not change five things at once:

Keep composition and subject identical.
Change ONLY: [one item, e.g. headline or clutter]
Do NOT change: [product angle, other text]

More: multi-turn editing blog · image editing tutorial.

Step 4: Ratio check and export

Verify critical copy, then export. For many variants: batch tutorial.

Step 5: Save a personal DNA block

[My DNA]
Brand color: #0EA5E9
Avoid: cluttered sticker walls, wrong logos
Default ratio: 1:1

Paste DNA before scene prompts next time.

FAQ

Slow generation? Thinking often adds tens of seconds; use Instant for rough drafts.
Broken CJK titles? Shorter quoted text + Thinking.
What next? Prompt engineering · Image 2.5 guide · Batch.

Done checklist

• Copied a library prompt
• Generated ≥1 image in the workbench
• Completed ≥1 single-dimension edit
• Exported at target ratio
• Saved a personal DNA draft

You now have the core GPTImage Prompts loop—deepen by scene blogs and tutorials next.