higgsfield-image

Higgsfield image generation (higgsfield-image)

Overview

This skill calls the image generation tool of Higgsfield MCP. It extracts the topic, purpose, and audience from the user's natural-language request, selects the most suitable one from the 11 official image models, and generates automatically.

Trigger keywords

Higgsfield image, Soul, Soul 2.0, Soul Cinema, Nano Banana, Nano Banana Pro, GPT Image, GPT Image 2, Seedream 4.0, Flux Kontext, Wan 2.2 Image, Wan 2.5, AI image generation, cinematic image, character consistency, 4K image

Prerequisite — MCP registration

moai-media When installing the plugin, .mcp.jsonautomatically registers this Higgsfield MCP.

{
  "higgsfield": {
    "type": "http",
    "url": "https://mcp.higgsfield.ai/mcp"
  }
}

Immediately before the first call, OAuth authentication must be completed once in Cowork (seemoai-media/CONNECTORS.md ).

11 official image models — at a glance

higgsfield.ai Based on the official page (2026-05).

Soul series (Higgsfield's own) ★

Model	Strengths	Best for
Soul	Cinematic grading and portrait detail (1st generation)	General cinematic
Soul 2.0	Improved version of Soul 1, enhanced detail and expressiveness	Recommended for portrait series and new work
Soul Cinema	Specialized in movie looks, 35mm and film grading	Advertising key visuals and video thumbnails

With the Soul Characters feature, learns character references → series consistency.

Nano Banana series (based on Gemini 3)

Model	Strengths	Best for
Nano Banana	Text rendering and basic quality	General social media and infographics
Nano Banana Pro	2K+ resolution and general card tone	General social media cards and print support

GPT Image series (OpenAI)

Model	Strengths	Best for
GPT Image	OpenAI 1st-generation image model	General purpose
GPT Image 2 ★	Improved OpenAI version, Text rendering SOTA·Factuality and accuracy ↑	Card news and posters ★·Realistic illustrations and products

Seedream

Model	Strengths	Best for
Seedream 4.0	Various styles, experimental tones, and artwork	Marketing visuals and art

Flux series

Model	Strengths	Best for
Flux Kontext	Photo-like realism and refined materials and lighting	Product shots, landscapes, and interiors

Wan series (latest from China)

Model	Strengths	Best for
Wan 2.2 Image	Various tones and experimental expression	Diversifying marketing tones
Wan 2.5	Wan 2.2 improved version	Latest tone

Detailed per-model usage patterns are references/model-guide.md.

Workflow

Step 1 — Identify intent

Extract the following from the user's one-line request:

• Topic (e.g., "30-year-old Korean woman looking at a laptop in a meeting room")
• Purpose (marketing key visual · SNS card · product shot · character series · presentation slide)
• Audience (B2B executives · general public · designers · developers)
• Tone (cinematic · minimal · dark · warm · trustworthy)

If insufficient, 1 round of AskUserQuestion.

Step 2 — Automatic model selection

Keyword matching. If the match is ambiguous, present 2-3 candidates.

User expression	Automatic selection
"Accurate text", "poster", "card news"	GPT Image 2 (1st priority) or Nano Banana Pro (2K)
"Cinematic", "film look"	Soul Cinema or Soul 2.0
"Character detail", "character"	Soul 2.0 (+ Soul Characters reference)
"Photorealistic", "realistic", "product shot"	Flux Kontext (1st priority) or GPT Image 2
"Artistic", "unique tone", "artwork"	Seedream 4.0
"Experimental", "new tone"	Wan 2.5
"General-purpose · fast"	GPT Image or Nano Banana

Step 3 — Parameter design

Parameter	값	Recommended default
prompt	Text description (English · Korean OK)	User request + per-model tone adjustment
width_and_height	1696x960(16:9)·1152x2048(9:16)·2048x1536(4:3)·1536x2048(3:4)·1024x1024(1:1)	Automatic by purpose
quality	720p / 1080p / 2K / 4K	General 1080p · Print 2K·4K
batch_size	1 / 4	1 (fixed) / 4 (exploration · A/B)
enhance_prompt	true / false	true (default)
style_id	Style preset UUID	Automatic if none
style_strength	0.0-1.0	1.0
seed	1-1000000	Unspecified (variation) / fixed (reproduction)
custom_reference_id	Soul Characters reference UUID	Series consistency
image_reference_url	Reference image URL	Option

Step 4 — Automatic aspect ratio mapping

Purpose	Aspect ratio	width_and_height
Instagram feed	1:1	1024x1024
Instagram story · reels	9:16	1152x2048
Facebook · Twitter · blog	16:9	1696x960
Print · poster (portrait)	3:4	1536x2048
Print · poster (landscape)	4:3	2048x1536

Step 5 — MCP call

ToolSearch(query: "select:mcp__higgsfield__generate_image")

mcp__higgsfield__generate_image({
  prompt: "[모델별 톤 보정된 프롬프트]",
  model: "soul_2_0",  // 또는 "nano_banana_pro" 등
  width_and_height: "1696x960",
  quality: "1080p",
  enhance_prompt: true,
  batch_size: 1
})

Model identifiers use the underscore_separated format on the official page (e.g., soul_2_0, nano_banana_pro, seedream_4_0, flux_kontext).

Step 6 — Asynchronous job polling

Higgsfield는 Asynchronous processing. Immediately after the call, receive a job ID and poll the status:

Status	Meaning
queued	Waiting (longer if balance is insufficient)
in_progress	Processing (image 5-15 seconds)
completed	Completed — result image URL in the field
failed	Failed (prompt · parameter issue)
nsfw	Content filtering

Step 7 — Review · variation

User response	Follow-up action
"Good"	Save · download
"Change this"	Same seed + partial edit
"Different direction"	Change seed + batch_size=4
"Photo in a different style"	Change model (Flux Kontext ↔ Soul 2.0, etc.)
"Same person, different pose"	Soul Characters reference UUID

Character consistency — Soul Characters

Official Higgsfield feature. Combined with Soul or Soul 2.0:

1. 첫 호출 (Soul 2.0): 원하는 인물 1장 생성
2. Soul Characters에 reference 등록 → UUID 수령
3. 이후 호출: custom_reference_id 파라미터에 UUID 전달
4. 동일 인물 + 다른 포즈·배경·표정 생성

Essential for marketing campaigns, brand characters, and series card news.

Usage examples

Example 1 — 4 images for card news (accurate text)

요청: "스타트업 투자 카드뉴스 4장 이미지 만들어 줘 (글자 들어감)"

자동 선택: GPT Image 2 (글자 렌더링 SOTA) — 1순위
대안: Nano Banana Pro (2K, 보조)
비율: 1024x1024 (인스타 피드)
quality: 2K
batch_size: 4

Example 2 — Cinematic brand key visual

요청: "한국인 30대 여성 CEO가 회의실에서 노트북 보는 시네마틱 샷"

자동 선택: Soul 2.0 (인물 디테일) 또는 Soul Cinema (영화 룩)
비율: 1696x960 (16:9)
quality: 1080p

Example 3 — Character series

1차: Soul 2.0 → 인물 1장 → Soul Characters에 reference 저장
2차+: 같은 reference UUID로 동일 인물 다양한 포즈

Example 4 — Product photo (realistic)

요청: "프리미엄 가죽 지갑 제품 샷, 흰 배경"

자동 선택: Flux Kontext (사진 사실성)
비율: 1024x1024
quality: 1080p

Example 5 — Ad key visual (cinematic tone)

요청: "신제품 광고 키 비주얼, 35mm 필름 룩"

자동 선택: Soul Cinema
비율: 1696x960
quality: 1080p

Output format

## Higgsfield 이미지 생성 결과

### 호출 정보
- 모델: [선택 모델, 예: soul_2_0]
- 프롬프트: [최종 프롬프트]
- 비율·해상도: [width_and_height · quality]
- Job ID: [Higgsfield job ID]

### 결과
- 이미지 URL: [Higgsfield CDN URL]
- 미리보기: ![](URL)
- 다운로드 경로: [로컬]

### 검수
- 텍스트 정확성·비율·해상도: [PASS]
- 의도 부합: [점수]

### 후속 추천
- 같은 seed로 부분 수정
- 다른 비율 변형
- Soul Characters reference 저장 (시리즈 작업 시)

Cost management

• Credit tiers differ by model (Soul Cinema/Pro tier costs more)
• batch_size=4costs 4x the credits
• quality=2K·4K가 1080pmore expensive than
• Workspace balance: higgsfield.ai → Billing

Precautions

Do

• Both English and Korean prompts are OK
• The first call is batch_size=4explore with, then finalize
• For character series, ensure consistency with Soul Characters reference
• For print/high quality, quality=2K or higher

Don't

• Prompts longer than 200 words
• Requesting contradictory elements at once ("minimal + flashy")
• NSFW or sensitive content
• Depictions of people that may infringe portrait rights
• Direct depiction of copyrighted characters or logos

Troubleshooting

Symptom	Cause	Solution
"Not connected"	OAuth unauthenticated	Cowork → Settings → MCP → Higgsfield → Connect
queued stuck	Insufficient balance	higgsfield.ai → Billing top-up
failed	Prompt or parameter error	Simplify prompt, use standard width_and_height
nsfw	Content filtering	Remove sensitive elements
Low result quality	Unsuitable model	Refer to the table and use a different model
Awkward Korean results	Model's Korean limitations	Retry with an English prompt

Related skills

Skill	Timing
moai-media:higgsfield-video	Follow-up: Turn images into video
moai-media:gemini-3-image-prompt	Alternative: Generate prompts only (external tools)
moai-media:gpt-image-2-prompt	Alternative: Use external ChatGPT
moai-content:card-news	Follow-up: Place images into card news
moai-design:claude-design-system-prep	Auxiliary: Finalize brand tone
moai-marketing:campaign-planner	Auxiliary: Campaign series images
moai-commerce:commerce-product-image-pipeline	Auxiliary: E-commerce batch

References

• references/model-guide.md — Strengths, weaknesses, and example prompts for 11 models
• Higgsfield official site — Model list (11 image + 11 video + 30+ ad)
• Higgsfield MCP guide
• Higgsfield Skills

Model list source

The list of 11 image models for this skill is higgsfield.ai official main page based on the stated criteria (2026-05). Models may be added or changed according to Higgsfield policy.