画像生成AIエージェント

Hunyuan Image 3.0による最新世代のテキストから画像生成。細部表現と多彩なアスペクト比を高精度に描写

FLUX.1 Kontext [max]による効率的な画像編集。単一の参照画像とテキストプロンプトを使用して高品質な画像変換を実現。

src:
  # Generate a base image
  base_image.jpg:
    agent: t2i-flux-schnell
    content: A beautiful landscape with mountains and lake
    dependency: []
    
  # Apply transformation with single dependency
  enhanced_landscape.jpg:
    agent: i2i-flux-max-kontext
    content: Transform into magical fantasy landscape with glowing elements
    dependency: ["src/base_image.jpg"]
    dependency_wait: true
    parameters:
      guidance_scale: 7.5
      aspect_ratio: "16:9"

  # Direct image URL usage
  stylized_photo.jpg:
    agent: i2i-flux-max-kontext
    content: Apply artistic oil painting style with vibrant colors
    image_url: https://example.com/photo.jpg
      safety_tolerance: "3"

Google Geminiベースの画像編集モデル。複数画像を同時に扱いながら高精度な編集を実現します。

src:
  images:
    base_cat.png:
      content: |-
        Simple cat image
        - Orange tabby cat sitting on white background
        - Facing forward
        - High resolution and clean image
      dependency: []
      agent: "t2i-flux-1"
      api: []
    edited_cat.png:
      content: |-
        Generate an image by adding a hat to the cat
        - Add a red hat to the original cat image
        - Place the hat naturally while maintaining the cat's characteristics
        - High-quality editing results
      dependency:
        - src/images/base_cat.png
      dependency_wait: true
      agent: "i2i-nano-banana-edit"
      api: []

ByteDance Seedream 4.0を用いた統合型の画像生成・編集。最大10枚の参照画像と高解像度出力に対応し、安全性チェックも切り替え可能です。

src:
  assets:
    base_cat.png:
      content: |-
        元画像用の猫の写真を生成してください
      dependency: []
      agent: "t2i-flux-1"
      api: []
    edited_cat_seedream.png:
      content: |-
        generate an image with a red hat on the cat
        - keep the cat's pose consistent
        - adjust background to a Victorian room
      dependency:
        - src/assets/base_cat.png
      dependency_wait: true
      agent: "i2i-bytedance-seedream-v4-edit"
      parameters:
        image_size:
          width: 1024
          height: 1024
        enable_safety_checker: true
        num_images: 1
      api: []

WAN 2.5 Previewによる画像編集モデル。最大4枚の参照画像を融合し、解像度やネガティブプロンプト、seedで細かな調整が可能です。

src:
  references:
    base_scene.png:
      content: |-
        Generate a cinematic cyberpunk alley image
      dependency: []
      agent: "t2i-flux-pro"
      api: []
    lighting_reference.png:
      content: |-
        Generate a dramatic lighting reference with blue and magenta neon highlights
      dependency: []
      agent: "t2i-flux-1"
      api: []
  edits:
    wan25_edited.png:
      content: |-
        Reimagine the alley with reflective puddles and a lone figure holding an umbrella
        - integrate the neon lighting reference for color mood
        - keep the original composition and camera angle
      dependency:
        - src/references/base_scene.png
        - src/references/lighting_reference.png
      dependency_wait: true
      agent: "i2i-wan-25-preview"
      parameters:
        image_size: "square_hd"
        negative_prompt: "blurry, low resolution, distorted, extra limbs"
        num_images: 1
        seed: 12345
      api: []