
そういえば今年のハロウィンは写真撮ってなかったので、画像生成AIでハロウィン画像を生成。Macローカルで、Qwen ImageやQwen Image Edit 2509など試してみました。

Qwen Imageでテキストから画像生成。ComfyUIでGGUFのモデルなど使えば問題なく使えました。

Prompt1:
A heartwarming Halloween parade of adorable kitten figurines designed like realistic felt toys. The kittens are marching in a festive autumn forest, each wearing tiny costumes — witch, ghost, pumpkin, bat, and fairy — holding small lanterns and candy baskets. Some kittens walk in pairs, others wave happily, surrounded by carved jack-o’-lanterns, colorful fallen leaves, and soft orange evening light. The scene is highly detailed, cozy, and magical, with a miniature diorama aesthetic, realistic felt toy texture, shallow depth of field, and cinematic composition.
Negative Prompt:
blurry, out of focus, low resolution, low quality, overexposed, underexposed, bad lighting, harsh shadows, deformed, mutated, extra limbs, missing limbs, distorted face, human-like, creepy, uncanny, doll with human features, plastic texture, flat lighting, inconsistent proportions, messy background, poor composition, noise, artifacts, watermark, text, signature, unrealistic fur, unnatural eyes, over-saturated colors, cheap toy look, CGI style, 3D render look, low detail, pixelated

Prompt2:
A dark Halloween parade of kitten figurines designed like realistic felt toys, marching through a misty forest under the moonlight. The kittens wear tiny witch, ghost, vampire, pumpkin, and skeleton costumes, holding flickering lanterns and eerie glowing pumpkins. The path is covered with fallen leaves and faint fog, lit by pale moonlight and candlelight from jack-o’-lanterns. Shadows stretch long, creating a mysterious and slightly spooky mood. The scene is a realistic miniature diorama with intricate toy textures, dramatic lighting, cinematic composition, and shallow depth of field.
Negative Promptは同じ。

Qwen Image Edit 2509では、入力画像を元にプロンプトで編集とかできますね。

Prompt1の画像をPrompt2のプロンプトで編集すると、構成は同じで雰囲気だけ変えることができました。



