Posters — event KV / e-commerce main image / Chinese text

What's the #1 complaint in designer Slack channels? It's not Photoshop lag. It's Chinese headline text.

Here's the standard Midjourney-era workflow for an Event KV poster:

1. Midjourney spits out the base image (30 seconds)
2. Pull it into Photoshop, mask it out, clean the edges (10 minutes)
3. Type the Chinese headline, realize the font isn't installed, download Source Han Serif (5 minutes)
4. Tweak kerning, leading, color, drop shadows (10 minutes)
5. Client says "make it bigger / change the color / try another version" (10 minutes × N rounds)
6. Export. Send to review. Get revisions. Loop.

A single poster from zero to live: 45 minutes to 2 hours. Five Event KVs a week eats two full work days.

gpt-image-2 just deleted the "Chinese text" step. 99% character-level accuracy means you write one prompt, the headline lands on the image, it's correct, and the typographic vibe matches what you asked for.

This chapter ships full prompt templates for three poster types (Event KV / E-commerce main image / Course cover), plus four real-world failure modes. After you finish, you can ship a publishable poster in 6 minutes.

---

1. Key Dimensions for Three Poster Types

Different distribution channels want different ratios. The first line of your prompt has to spell this out:

⚠️ Print posters need 300 DPI (gpt-image-2 defaults to 72 DPI for screen). Use Photoshop's "Image Size" to bump DPI without resampling pixels.

---

2. Event KV Prompt Template

The core of an Event KV: the visual carries emotion, the headline carries information, the negative space carries the product/logo.

Full template:

Wide horizontal banner, 16:9 ratio, 1920×1080.

Background: a young diverse team in a modern office,
laptops open, brainstorming session, warm afternoon light
from large windows on the right side.

Foreground text overlay (left half, occupies left 50%):
- Headline (top, huge bold red Chinese, centered):
  "AI 训练营"
- Subhead (below headline, medium white):
  "30 天从 0 到部署一个 AI 应用"
- Footer (bottom, small dark gray):
  "2026.05.20 - 06.20  |  线上直播"

Right half (50%): clean office background, person silhouette,
leave subtle empty space for company logo placement (top-right corner, 200×100 area).

Style: editorial corporate photography, slight film grain,
warm color grade, high contrast text outlined for readability.

Exact text only. No duplicate. No watermarks.

Breakdown

• Line 1: Dimensions first, so the ratio doesn't drift
• Background block: Scene + people + lighting (the more specific, the more stable)
• Text layer: Each line needs role hint + double quotes + position + font size + color — all 5, no skipping
• Negative space block: Tell the model explicitly to leave a clean area for the logo
• Style block: Words like "editorial" or "corporate" hold up better in LinkedIn / business contexts
• Constraint block: Three sentences kill 90% of the failures

---

3. E-commerce Main Image Prompt Template

E-commerce main images are pickier than Event KVs: one image has to sell the product AND the selling points.

Square product photography, 1:1, 800×800.

Subject: a sleek black wireless earphone case, 3/4 angle,
floating slightly above a marble surface,
soft contact shadow below product, perfectly clean lighting.

Top-right corner large red badge text (in a circular badge):
"限时 5 折"

Bottom horizontal Chinese tagline (centered, dark gray, medium):
"续航 36 小时 · 主动降噪 · 蓝牙 5.4"

Style: clean studio softbox lighting, white seamless background,
high commercial polish, subtle product reflection, e-commerce ready.

No extra text, no duplicate, exact Chinese text only.
Top center is reserved (empty) for platform price overlay.

Three E-commerce-Specific Tricks

1. Pure white background: Spell out white seamless background, otherwise you get a faint gray shadow
2. Floating product + shadow: floating slightly above + soft contact shadow makes the product feel weightless
3. Reserve the price zone: Taobao / JD overlay prices and coupons automatically, so the top-center 200×100 zone has to stay clean

---

4. Course Cover Prompt Template

The core of a course cover: earn the click first, then deliver the value.

Course cover, 16:9, 1280×720.

Left half (60%): a developer's hands on a glowing keyboard,
neon blue + electric purple light bleed, cyberpunk aesthetic,
steam rising from a coffee cup foreground (slight blur).

Right half (40%, dark navy clean background):
Three-line text stack, vertically centered:
- Top (huge bold yellow): "Vibe Coding 训练营"
- Middle (medium white): "用 Cursor + Claude Code 30 天交付"
- Bottom (small gray): "Web 全栈方向 · 2026 春季 · 限招 30 人"

Style: cinematic editorial tech aesthetic, sharp contrast,
high commercial polish, attention-grabbing but professional.

Exact text only, no decorative elements interfering with text area.

---

5. Real-World Failures

Failure 1: Chinese Text "Looks Right but Isn't"

First time I made a KV for an AI bootcamp, the headline was "AI 工程师训练营". Came out perfect. Then the client zoomed in and noticed the "川" radical on the right of "训" was missing a stroke.

It's easy to miss with a quick scan. You have to zoom to 200% and check every single character.

Fix:

• After generating any high-quality poster, zoom to 200% and proof every character
• Avoid characters with dense strokes ("嬴", "龘", "鬱"), or swap in a simpler synonym
• The bulletproof fallback: write no text in the base image prompt and add the text layer in Photopea / Figma afterward

Failure 2: Brand Color Is Off

Write bright red in the prompt and you'll get orange-red, peach-red, dark-red — 8 images, 8 different reds. Client says: "Our brand color is #FF5757, not whatever this is."

Fix: Use hex codes, not color adjectives.

• ❌ bright red / vibrant blue
• ✅ red #FF5757 / blue #1B4DFF
• For large color blocks, drop the hex right in: background color #FFF5E6

Failure 3: Logo Zone Gets Filled In

Vague instructions like leave space for logo get ignored constantly. You generate the image, and there's nowhere to put the logo — it'll cover either the subject or the text.

Fix: You need explicit coordinates.

• ❌ leave space for logo on top-left
• ✅ leave the top-left 200×100 pixel area completely clean (white space) for logo placement
• You can also describe a virtual frame: reserve a clean rectangular area in top-left corner, approximately 25% × 14% of the image, for logo overlay

Failure 4: Print Bleed

You print an offline poster and either there's a 3mm white border around the edges, or critical content gets sliced off. Why: screen posters bleed to the edge, but print posters need an extra 3mm bleed area outside the trim line.

Fix: For print posters, add with 3mm bleed area on all sides, key content within safe area inset 5mm from edges to the prompt. Or add the bleed in Photoshop afterward.

---

6. Real Case Study from JR Academy

In 2026-04 we made spring Event KV posters for 8 bootcamps, 5 posters each:

That's about 55 hours saved — roughly 1.5 weeks of a full-time designer's work.

But there's an upfront cost: each bootcamp needs a "style anchor" defined first (typical scene + color palette + typographic feel), or all 5 posters end up tonally inconsistent. We later stored each bootcamp's "style anchor prompt" in a shared Notion page, so ops folks can swap the headline text and batch-generate.

---

7. What's Next

The next chapter dives into Chinese social media: Ch 08 Xiaohongshu takes this poster skillset to the extreme — vertical covers + Chinese headline rendering at maximum difficulty.

If you need to ship a poster right now:

1. Pick one of the three templates from §2 / §3 / §4
2. Swap the background description + headline text + brand hex
3. Generate 8 at once, pick 1
4. Run through §5's four failure modes one by one
5. For complex Chinese characters, zoom to 200% and proof. If something's off, fix the text in Photopea afterward

Posters are the first scene where gpt-image-2 truly flips the table. Once you've made one perfect Event KV, going back to typing headlines in Photoshop feels like a waste of your life.

---

📷 Real Poster Examples

Three real-world poster examples, sourced from awesome-gpt-image (CC BY 4.0). You can paste these prompts straight into ChatGPT to reproduce — change two words and you've got a new theme.

Case 1: 1980s Propaganda Poster Style (Chinese slogan + multiple characters)

Prompt:

Generate a 1980s propaganda poster. Use the exact slogan "热烈庆祝GPT-Image-2全量开放". Include Sam Altman, Dario Amodei, and Elon Musk, and give Dario Amodei a red scarf.

Notice how this prompt nails down four things at once: style + Chinese slogan + characters + decorative detail. Even "give Dario the red scarf" is spelled out. The generation lands on the first try. This is the inverse proof of §5's Failure 1 (Chinese text "looks right but isn't") and Failure 3 (logo zone): write the constraints out, and the model gives you exactly what you asked for.

📷 Creator: @akokoi1 · Featured in: awesome-gpt-image

Case 2: Celebrity Endorsement Car Poster (short prompt + 4 same-theme variants)

Prompt:

XXX 代言 XXX 汽车海报

Just swap the two XXXs (celebrity + car model) and you can mass-produce "celebrity endorsement" posters. The model fills in the background, composition, text, and brand vibe by itself. This is the poster child for short-prompt Chinese + strong scene understanding — you can use the same prompt to generate 30 different celebrity × car combos in 5 minutes.

📷 Creator: @johnAGI168 · Featured in: awesome-gpt-image

Case 3: E-commerce Product Detail Page (4 product types)

Prompt:

Generate an e-commerce product detail page for [product name]

The standard product detail page structure (main image + selling points + price + specs + reviews) — gpt-image-2 nails it in one sentence. Maps directly onto Taobao / JD detail page styling. Swap the product name, reuse the prompt, batch-generate. This is the throughput multiplier for §3's e-commerce main image chapter.

📷 Creator: @MrLarus · Featured in: awesome-gpt-image