last update
May 27, 2026
license
MITv1.0.0
format
SKILL.md
- Install
// Downloads SKILL.md into ~/.claude/skills/midjourney-prompt-pro/
$ curl -fsSL --create-dirs -o ~/.claude/skills/midjourney-prompt-pro/SKILL.md https://claudeskil.com/skills/midjourney-prompt-pro/SKILL.md - Invoke
// Inside Claude Code, type the skill name as a slash command
> /midjourney-prompt-pro help me ship this - Or just ask
// Claude loads the skill by itself when a request matches its description
> help me ship this
- name:
- Midjourney Prompt Pro
- slug:
- midjourney-prompt-pro
- version:
- v1.0.0
- license:
- MIT
- author:
- @chiragthummar16
- categories:
- tags:
- #midjourney#midjourney-v7#image-ai#prompt-engineering#style-ref
- description:
Midjourney v7 prompts: style references, aspect ratios, weight syntax, the things that move quality 2× up.
// What you can do with it
- Style references via --sref with --sw (style weight, 0–1000) to dial influence up or down.
- Character/subject consistency via --cref plus --cw (0–100) — high for face+hair+clothes, low when you only want the face held.
- Aspect ratios mapped to use case: --ar 16:9 for product/hero, 4:5 for paid social, 9:16 for reels and stories, 1:1 for catalog tiles.
- Stylization and variety with --stylize (--s) and --weird, and prompt-internal token weighting using :: (e.g. red car::2 snow::0.5).
- Negatives with --no, which works on concrete nouns (--no text, watermark, hands) far better than vague adjectives.
README.md
Midjourney Prompt Pro
A prompt-engineering assistant tuned specifically for Midjourney v7. It writes, debugs, and version-controls prompts using v7's real parameter grammar — --sref, --cref, weight modifiers, aspect ratios, and stylization — so you stop guessing why one render looks great and the next looks nothing like it. Built for designers, marketers, and anyone shipping image batches who needs results that reproduce across reruns.
When to use it
- You found a look you love and need it to survive across a whole campaign, not vanish on the next
/imagine. - You're producing the same asset in multiple crops (hero, ad, reel) and want one coherent prompt family.
- A render keeps drifting — wrong subject, wrong palette, mystery artifacts — and you need to know which token is fighting you.
- You're onboarding a team and want a documented prompt convention instead of tribal Discord lore.
What it does
The skill knows v7's flag syntax and applies it deliberately:
- Style references via
--sref <url or code>with--sw(style weight, 0–1000) to dial influence up or down. - Character/subject consistency via
--cref <url>plus--cw(0–100) — high for face+hair+clothes, low when you only want the face held. - Aspect ratios mapped to use case:
--ar 16:9for product/hero,4:5for paid social,9:16for reels and stories,1:1for catalog tiles. - Stylization and variety with
--stylize(--s) and--weird, and prompt-internal token weighting using::(e.g.red car::2 snow::0.5). - Negatives with
--no, which works on concrete nouns (--no text, watermark, hands) far better than vague adjectives.
It also explains tradeoffs rather than dumping flags — e.g. why stacking --sref + high --stylize washes out your subject.
Usage
curl -fsSL --create-dirs -o ~/.claude/skills/midjourney-prompt-pro/SKILL.md https://claudeskil.com/skills/midjourney-prompt-pro/SKILL.md
> Build me a v7 prompt: minimalist matte skincare bottle, soft studio light,
for a paid Instagram ad. Lock the style across a 6-image set.
# Returns:
/imagine prompt: minimalist matte skincare bottle, soft diffused studio
lighting, beige seamless backdrop, product photography
--ar 4:5 --sref 1873402915 --sw 120 --stylize 250 --no text, hands, reflection
Reuse that --sref seed across the set to keep every crop on-brand.
Tips & gotchas
- A numeric
--srefseed is far more reproducible than an image URL — image hosts expire, codes don't. --crefis for characters, not style. Using it to copy a vibe gives muddy results — that's--sref's job.--cw 0keeps only the face; bump toward 100 to lock outfit and pose too.--noignores prompt phrasing like "without people" — only the flag reliably removes things.
FAQ
Does it support v6 or niji? It's optimized for v7 syntax. Most flags carry back to v6, but --sw/--cw behavior differs, so specify your version.
Can it read an existing prompt and fix it? Yes — paste a prompt that's misbehaving and it'll flag the conflicting tokens and weights.
Will the same prompt give identical images? No — Midjourney is stochastic. Pinning --sref and --seed gets you consistent style and composition, not pixel-identical output.
$ cat reviews/