← 返回 Skill 列表
extension
分类: 内容与媒体API Key 暂未确认

midjourney

Midjourney AI图像生成。用于创意AI。

person作者: jakexiaohubgithub

Midjourney

Midjourney is an advanced generative image platform recognized for artistic composition, fine detail, photorealism, and parameter-driven prompt manipulation.

When to Use

  • High-End Concept Art & Creative Exploration: Synthesizing cinematic illustrations, character design, and game environments.
  • Photorealistic Architectural & Product Visuals: Generating high-fidelity mockups with precise lighting and texture prompts.
  • Style Consistency with Style References: Replicating specific artistic styles and palettes across multiple prompts using --sref.
  • Multi-Prompt Weighting & Composition: Blending concepts with explicit semantic weights using :: syntax.

Quick Start

/imagine prompt: cinematic wide shot of a futuristic data center with neon liquid cooling, architectural digest style, volumetric lighting, unreal engine 5 render --ar 16:9 --v 6.1 --style raw

Core Concepts

Core Command Parameters (v6 & v6.1)

Directing aspect ratio, stylization, and rendering engines:

/imagine prompt: cinematic photography of a sleek glass server rack in a minimalist data center, volumetric teal lighting, shallow depth of field, architectural digest style --ar 16:9 --v 6.1 --style raw --stylize 250 --quality 1

Parameter breakdown:

  • --ar 16:9: Sets wide landscape aspect ratio (9:16 for mobile, 1:1 for square).
  • --v 6.1: Specifies the Midjourney v6.1 rendering algorithm.
  • --style raw: Reduces Midjourney's default aesthetic bias for more photographic realism.
  • --stylize 250 (--s): Controls strength of artistic flair (range 0 to 1000).
  • --chaos 15 (--c): Adds variation to initial grid generations (range 0 to 100).

Style References (--sref) & Character Consistency (--cref)

Transferring aesthetic signatures across scenes:

# Style Reference: match aesthetic of an existing reference image
/imagine prompt: an autonomous delivery drone flying over a futuristic metropolis at sunset --sref https://example.com/style_sample.jpg --sw 800

# Character Reference: maintain subject identity
/imagine prompt: the detective sitting at a rainy cafe table reviewing case files --cref https://example.com/detective_face.jpg --cw 90

Multi-Prompting with Explicit Weights (::)

Preventing concept bleed and tuning semantic emphasis:

# Concept separation: 'hot' and 'dog' instead of a frankfurter
/imagine prompt: hot::2 dog::1 running on beach --ar 3:2

# Negative weighting: suppress specific elements without prompt negative words
/imagine prompt: vibrant Tokyo street market at night, cinematic neon reflections --no blur, text, watermark, vehicles

Common Patterns

Character Consistency with Style and Character Reference (--sref, --cref)

Problem: Generating the same character or style across multiple scenes produces completely different visuals.

Solution: Use --cref (character reference) and --sref (style reference) parameters:

/imagine prompt: cybernetic detective walking down a rainy alley, night time, cinematic lighting --cref https://url-to-character-face.png --cw 80 --v 6.1

/imagine prompt: cozy cafe interior with warm morning light --sref https://url-to-style-reference.png --sw 100 --v 6.1

Best Practices

Do:

  • Use --style raw when aiming for accurate photorealism and literal prompt adherence.
  • Specify lighting, camera lenses, and film stock (e.g. 35mm lens, f/1.8, golden hour illumination) for realism.
  • Leverage --sref (Style Reference) with --sw (weight) to maintain brand aesthetic across a series of images.
  • Use --no for negative conditions rather than writing phrases like "without people" in the main prompt.

Don't:

  • Use buzzwords like "photorealistic", "hyperrealistic", or "4K"; describe physical lighting and textures instead.
  • Use long, rambling paragraphs; concise, comma-separated descriptive descriptors yield better results.
  • Exceed --stylize 750 unless abstract, highly stylized, or artistic interpretation is desired.

Troubleshooting

| Error | Cause | Solution | | :--------------------------------------- | :------------------------------------------------------------------------------ | :------------------------------------------------------------------- | | Invalid parameter: unrecognized flag | Typo in parameter name or space between double dashes. | Ensure format is exactly --ar 16:9, --v 6.1, or --stylize 250. | | Image reference URL cannot be accessed | Direct link to image URL is private, expired, or not ending in image extension. | Ensure image URL is public and ends in .png, .jpg, or .webp. | | Banned prompt / Content filter trigger | Prompt contains words flagged by Midjourney community guidelines. | Rephrase prompt using descriptive, non-violent, non-explicit terms. |

References