# AI thumbnails + ad creative

> Midjourney owns the AI-image quality bar for marketing visuals. The bottleneck isn't the model — it's the prompt. Use Claude to turn a creative brief into a tight Midjourney prompt; render in Midjourney; finish with Canva for typography. The whole loop is 6 to 8 minutes per image.

Source: https://aistack.sh/stack/ai-thumbnails
Category: Marketing · Visual
Level: Beginner
Last verified by a human: 2026-05-05

## The tools

### Claude — Prompt engineer

- Pricing: Free · $20/mo Pro · $100/mo Max 5x · $200/mo Max 20x · Sonnet API $2/$10 per M tokens (input/output)
- Site: https://claude.ai
- Why this one: A great Midjourney prompt is the difference between brand-grade and stock-looking. Claude turns a 1-line brief into a structured Midjourney prompt with style, mood, lighting, lens, aspect ratio, and the right --ar / --v flags.
- Swap in instead: ChatGPT
- More: https://aistack.sh/tool/claude

### Midjourney — Image generation

- Pricing: $10/mo Basic · $30/mo Standard · $60/mo Pro · $120/mo Mega
- Site: https://midjourney.com
- Why this one: Frontier-quality output with the most consistent house style. Worth the $30/mo Standard for unlimited Relax-mode generation.
- More: https://aistack.sh/tool/midjourney

### Canva — Typography + brand polish

- Pricing: Free · $15/mo Pro
- Site: https://canva.com
- Why this one: Midjourney can't render text reliably. Canva's templates handle the headline, the logo, the CTA — the parts that need pixel-precision and brand consistency.
- Swap in instead: Figma
- More: https://aistack.sh/tool/canva

## Monthly cost

### $30/mo — Hobby (small)

- Claude: Free tier
- Midjourney: $10 Basic
- Canva: Free + $20 ($240/yr Pro annualized)

### $50/mo — Daily creator output (medium)

- Claude: $20
- Midjourney: $30 Standard
- Canva: Free

### $95/mo — Brand or agency (heavy)

- Claude: $20
- Midjourney: $60 Pro
- Canva: $15 Pro

## Workflow

### 1. Brief to Midjourney prompt (Claude)

One Claude turn. Brief in, structured prompt out.

**Prompt: Creative brief → Midjourney prompt**

```
Turn the creative brief below into a Midjourney prompt that produces a brand-grade image. Don't generate, just write the prompt.

Brief:
"""
{{1 to 3 sentences: what's the image for, what should it convey, brand mood, where it'll be used (thumbnail / ad / OG image / social)}}
"""

Output the prompt in this exact structure (one line):

[subject], [composition], [style + reference artist or movement], [lighting], [color palette], [lens / camera if photo-real], [mood], --ar [aspect ratio], --v [latest version], [optional: --style raw or --niji]

Constraints:
- Specific, not generic. Name an artist, a film, or a concrete reference.
- One mood, one lighting direction. Don't list 5 contradictory adjectives.
- For thumbnails: --ar 16:9. For OG images: --ar 1.91:1. For Reels: --ar 9:16.
- Strip "ultra-realistic 8k masterpiece" filler — adds nothing to current Midjourney.
- For brand work, prefer --style raw (less stylized).

Output the prompt only. No commentary.
```

### 2. Generate in Midjourney (Midjourney)

Paste prompt. Generate 4 variants. Pick the strongest, run /upscale.

### 3. Polish in Canva (Canva)

Drop the upscaled image into a Canva template. Add headline + logo + CTA. Export at the platform's native size.

### 4. QA on the device (Canva)

Open the final asset on the actual device + platform you're posting to. Mobile thumbnails legibility is the most-missed step.

## What it produced

**YouTube creator, thumbnail CTR lift**

Switched from Photoshop + stock to this stack. Average thumbnail CTR rose from 4.6% to 7.1% across the next 30 videos. Per-thumbnail time dropped from ~25 min to ~7 min.

## Pitfalls

- **Trusting AI to render text** — Midjourney still botches text in most images. Generate the visual; add text in Canva.
- **Relax mode at brand speed** — Midjourney Relax mode queues your generations; can take minutes during peak hours. Pay for Fast hours when you're shipping under deadline.
- **Brand drift** — Without a locked palette + style reference, Midjourney happily generates 4 'similar' images that don't match each other. Save your prompt + seed; reuse them.

---

Curated by tone-d on aistack.sh. Last updated 2026-05-05.