moai-coworker | image prompt builder (text output only)
Overview
Midjourney v8 Alpha (released 2026.03.17) and v8.1 (announced 2026.03.21) introduced 5x faster generation, native 2K (--hd), extra coherence (--q 4), improved text rendering, --stylize recommended up to 1000, and a new --sv 7 Style Reference·Moodboards model. This skill converts a user's one-line request into a structure based on the official Midjourney Parameter List:
[subject], [scene keywords], [composition], [lighting], [style]
--ar W:H --style raw --hd --q 4 --s 0~1000
[--sref CODE|URL --sw N] [--oref URL --cw 0~100]
[--p PROFILE] [--no NEGATIVE] [--c 0~100]
In particular, this skill:
- 3-model simultaneous output: Midjourney main + GPT-image-2 (6-Block) + Gemini 3 Pro Image (5-component)
- Presets + fine-tuning: 4 presets × 4 slots
- Automated pitfall warnings:
--cw 100default value pitfalls,--hd+--q 4+--sref= 4x cost,--crefdeprecated - Personalization integration:
--p PROFILE_IDor--profile PROFILE_IDguidance - Style Code library:
--sref CODEreference to frequently used codes
Trigger keywords
Midjourney prompt MJ prompt Midjourney prompt imagine prompt Midjourney v8 Midjourney 8.1 sref prompt oref prompt style code prompt personalization profile
Workflow
사용자 자연어 한 줄
↓
[Round 1] AskUserQuestion — 프리셋 선택
↓
[Round 2] AskUserQuestion — 프리셋별 미세조정 (3~4 슬롯)
↓
[Round 3] AskUserQuestion — 화면비 + 텍스트 + 고급 옵션(--sref·--oref·--p·--no)
↓
[내부] 슬롯 → 키워드 콤마 + --파라미터 매핑
↓
[내부] 같은 슬롯 → GPT 6-Block + Gemini 5-component 변환
↓
[내부] 함정 검사 (--cw 100·4x cost·v8 비호환 옵션)
↓
출력: 3개 모델 프롬프트 + 비용 경고 + 한국어 해설
Execution rules
Round 1 — Preset selection (required)
Choose from 4 presets (product shot, portrait, illustration, landscape). The preset slot definitions use a single source shared by the 3 image prompt builders (gpt-image-2, gemini, midjourney), and the source is in the media-gpt-image-2-prompt skill:
- Product shot —
${CLAUDE_PLUGIN_ROOT}/skills/media-gpt-image-2-prompt/presets/product-shot.md - Portrait·character —
${CLAUDE_PLUGIN_ROOT}/skills/media-gpt-image-2-prompt/presets/portrait.md - Illustration·art —
${CLAUDE_PLUGIN_ROOT}/skills/media-gpt-image-2-prompt/presets/illustration.md - Landscape·environment —
${CLAUDE_PLUGIN_ROOT}/skills/media-gpt-image-2-prompt/presets/landscape.md
Round 2 — Preset-specific fine-tuning (3-4 questions)
Follows the slot definitions of the above shared slot source (${CLAUDE_PLUGIN_ROOT}/skills/media-gpt-image-2-prompt/presets/<name>.md). Each preset file contains tone conversion guides for all three models—GPT, Gemini, and Midjourney—and this skill applies the Midjourney keyword+parameter tone section.
Round 3 — Aspect ratio + text + advanced options
Q1 — Aspect ratio (--ar)
| Aspect ratio | --ar |
Purpose |
|---|---|---|
| 1:1 (recommended) | --ar 1:1 |
SNS square |
| 16:9 | --ar 16:9 |
Wide, YouTube |
| 9:16 | --ar 9:16 |
Reels·Shorts |
| 4:5 | --ar 4:5 |
Instagram feed |
| 3:2 | --ar 3:2 |
Photo standard |
| 2:3 | --ar 2:3 |
Book cover |
Q2 — Quality·style (--hd, --q, --style, --s)
| Option | Action | Cost |
|---|---|---|
| Default | Standard quality·auto style | 1x |
--style raw |
Natural color·photographic, minimal model intervention | 1x |
--hd |
native 2K | 4x |
--q 4 |
extra coherence | 4x |
--hd --q 4 |
둘 다 | 16x ⚠️ |
--s 0~1000 |
stylize (default 100) | Same |
Cost caution: --hd + --q 4 + --sref/moodboard combinations are 4x each Cost (multiplicative). Detailed pitfalls are in references/cost-traps.md.
Q3 — Text (when there is text in the image)
Midjourney v8 has improved text rendering compared to V6/V7, but it is not as accurate as GPT/Gemini. Recommended:
- Put text exactly in quotes:
"MONDAY" - Focus on short words/short sentences (long sentences risk breaking)
- English is more stable than Korean
Q4 — Advanced options (optional)
AskUserQuestionSelect additional options with:
| Option | Use | Note |
|---|---|---|
Style Reference (--sref) |
Specific style code or reference URL | 4x cost, --sw 0~1000 together |
Omni Reference (--oref) |
Character/object consistency | 2x cost, --cw 0~100 pitfall (default 100) |
Personalization (--p) |
User learning profile | Requires prior setup (starts at 40 ratings, 200 stable, 2000 max) |
Negative (--no) |
Elements to exclude | 예: --no people, --no text |
Chaos (--c 0~100) |
Result diversity | For drafts |
For detailed usage, see references/style-references.md.
Internal processing — keyword commas + --parameter mapping
[subject], [scene keywords], [composition keywords],
[lighting keywords], [style/medium keywords] --ar W:H
--style raw --hd --q 4 --s 250
references/prompt-blocks.mdRefer to the keyword classification guide in .
Internal processing — pitfall checks
This skill automatically checks the following before output and includes warnings in the Korean explanation:
--cwIf unspecified → "Default 100. If you want to bring only the face,--cw 20~40" guidance--hd --q 4 --srefcombination → "4x cost each, total 64x GPU time" warning- v8-incompatible options (
--cref, V6 specific, etc.) → automatically removed or deprecated note --sref random→ results cannot be reproduced, guidance to check the code
Output — 3 model code blocks
## 🎨 생성된 프롬프트 (3개 모델)
### 1) Midjourney v8.1 (메인)
, , , , --ar 1:1 --style raw --hd --q 4 --s 300
**비용 추정**: `--hd` 4x × `--q 4` 4x = 16x GPU 시간 (relax 모드 권장)
**Personalization 활용**: `--p YOUR_PROFILE_ID` 추가 시 일관된 스타일
### 2) GPT-image-2 — OpenAI ChatGPT / API
<6-Block natural language paragraph>
**권장 파라미터**: `quality=medium`, `size=1024x1024`
### 3) Gemini 3 Pro Image — Nano Banana Pro
<5-component English sentence>
**권장 파라미터**: `aspect_ratio=1:1`, `resolution=2K`, `mode=Thinking`
### 📝 한국어 해설
- Midjourney v8은 키워드 콤마 + `--파라미터` 어조에 최적화
- 텍스트 정확도: GPT/Gemini > MJ. 긴 글자는 GPT 권장
- `--cw 100` 함정: reference 이미지의 조명·스타일까지 상속됨
### 🔗 페어 스킬
- `media-gpt-image-2-prompt` — GPT 어조 (sibling)
- `media-gemini-3-image-prompt` — Gemini 어조 (sibling)
- Midjourney 실행은 Discord `/imagine` 또는 alpha.midjourney.com에서 직접
Usage examples
Example 1: Product shot
"Midjourney prompt, matte black mug 'MONDAY' text product shot, 1:1, --hd"
→ Round 1: product shot → Round 2: mug/slate/window/3-4 minutes → Round 3: 1:1 + "MONDAY" + --hd activated → 3 model outputs + 4x cost warning.
Example 2: Character consistency (--oref)
"Midjourney prompt for the same character in a different scene, --oref [URL] --cw 50"
→ person/character preset → Round 2 → in Round 3, --oref URL + --cw 50 input → in the output, --oref workflow Korean explanation + --cw guide.
Example 3: Style Reference
"MJ prompt --sref 1234567890 --sw 200"
→ Round 1·2 → in Round 3, enter sref code + sw → in the output, 4x cost warning + --sv 7 default note.
Output format
| Deliverable | Format | Description |
|---|---|---|
| Midjourney v8.1 prompt | Keyword commas + --파라미터 한 줄 |
Discord /imagine or enter at alpha.midjourney.com |
| GPT-image-2 prompt | English 6-Block natural language paragraph | Paste into ChatGPT / API |
| Gemini 3 Pro Image prompt | English 5-component paragraph | Paste into Google AI Studio / Vertex AI |
| Cost Estimate | Multiplier (e.g. 16x) | --hd·--q·--sref·--oref combination |
| Korean commentary | Markdown | Pitfalls·--cw·--sv·-personalization |
Precautions
- This skill only prompt text outputs. Actual image generation must be executed directly on Discord or alpha.midjourney.com (Midjourney official API automation restrictions).
--cref(V6 character consistency) is automatically replaced in V7·V8 with deprecated.--oref.--cw 100(default) Pitfall: the reference image's lighting and style are also inherited, conflicting with the new prompt. To take only the face, use--cw 20~40.--hd·--q 4·--sref/moodboard each 4x cost. When combined, costs multiply.--oref는 Fast Mode·Draft Mode·Conversational Mode·--q 4and incompatible.- In the early V8 Alpha, Relax Mode was not supported, but it was gradually added in V8.1. If cost-sensitive, check Relax availability.
--sref randommakes results non-reproducible. If you get a result you like, check and save the actual code.- Personalization (
--p) starts at 40 ratings, becomes stable at 200 ratings, and improves up to 2000 ratings. - All output images are used at the user's responsibility (copyright, portrait rights, brands).
Related Skills
| Skill | Relationship | Description |
|---|---|---|
| media-gpt-image-2-prompt | sibling | GPT 6-Block tone prompt from the same input |
| media-gemini-3-image-prompt | sibling | Gemini 5-component tone prompt from the same input |
| Higgsfield MCP (Soul) | alternative | API automatic generation (cinematic images and character unified into one, MJ not included) |
Sources
Primary recommended sources (official):
- Midjourney Documentation — Parameter List
- Midjourney Documentation — Style Reference (
--sref) - Midjourney Documentation — Omni Reference (
--oref) - Midjourney Documentation — Character Reference (deprecated
--cref) - Midjourney Documentation Hub
Industry references:
- Blake Crosley — Midjourney V8.1 + V7 Reference Guide
- AI Tools DevPro — Midjourney 2026: v8 Specs Manual
- ArtPromptHQ — Ultimate Midjourney Prompt Packs Guide
Based on the above sources, V8.1 parameters, --sref/--oref behavior, --cw pitfalls, the 4x cost matrix, Personalization rating stages, and V8-incompatible options were derived.