image-generation

图像生成

使用 each::sense 根据文本描述生成图像。each::sense 是一款智能 AI Agent，可自动为你的请求选择最佳模型。

快速开始

需要 each::labs API 密钥。请前往 eachlabs.ai 获取。

使用 curl

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -d '{
    "messages": [{"role": "user", "content": "A lone astronaut standing on a red desert planet, looking up at a sky filled with three massive moons, cinematic lighting, photorealistic, wide shot"}],
    "stream": false
  }'

使用 Python（OpenAI SDK）

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_EACHLABS_API_KEY",
    base_url="https://eachsense-agent.core.eachlabs.run/v1"
)

response = client.chat.completions.create(
    model="eachsense/beta",
    messages=[{"role": "user", "content": "A lone astronaut standing on a red desert planet, looking up at a sky filled with three massive moons, cinematic lighting, photorealistic, wide shot"}]
)

print(response.choices[0].message.content)

使用参考图像

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -d '{
    "messages": [{"role": "user", "content": [
              {"type": "text", "text": "Generate an image in the same style as this reference, but change the setting to a futuristic Tokyo street at night"},
              {"type": "image_url", "image_url": {"url": "https://example.com/reference-style.jpg"}}
            ]
          }
    ],
    "stream": false
  }'

图像会使用 OpenAI 多模态内容格式嵌入消息中发送。每个请求最多可包含 4 张图像。

流式传输

设置 "stream": true 以获取实时 SSE 响应，或设置 "stream": false" 以在单个响应中获取完整结果。流式传输适合在 UI 中显示进度；非流式传输则更适用于脚本和自动化。

可用模型

模型	优势	最适合
flux-2-max	最高品质，细节出众	主视觉图、可直接印刷的视觉作品、作品集创作
flux-2-pro	速度快且品质均衡	快速迭代、生产工作流
flux-kontext-pro	图像中的文字渲染准确	海报、标牌、带有清晰可读文字的徽标
nano-banana-pro	生成速度最快	原型设计、批量生成、实时预览
gemini-imagen-4	出色的照片级真实感	产品摄影、生活方式照片、真实场景
seedream-v4-5	艺术化和风格化的输出	插画、奇幻艺术、绘画风格
kling-text-to-image	适用于多种风格	通用图像生成、混合媒介

each::sense 会根据你的提示词自动选择最优模型。除非你想覆盖自动选择结果，否则无需指定模型。

提示词工程指南

提示词结构

[subject] + [action/pose] + [environment] + [style] + [lighting] + [camera/composition]

风格关键词

风格	关键词
照片级真实感	photorealistic, ultra-realistic, DSLR photo, 8K, raw photo, unedited
插画	digital illustration, vector art, hand-drawn, watercolor, ink drawing
电影感	cinematic, film still, movie scene, dramatic lighting, anamorphic lens
动漫	anime style, manga, cel shading, Studio Ghibli, Makoto Shinkai
3D 渲染	3D render, octane render, unreal engine, isometric, blender render
油画	oil painting, impasto, classical art, Renaissance, Baroque
像素艺术	pixel art, 16-bit, retro game, 8-bit, sprite art
极简主义	minimalist, flat design, simple, clean lines, geometric

宽高比

在提示词中指定宽高比，以控制图像尺寸：

"... landscape format, 16:9 aspect ratio"
"... square format, 1:1"
"... portrait format, 9:16, vertical"
"... ultrawide, 21:9 cinematic"

提升品质的关键词

masterpiece, highly detailed, sharp focus, professional,
studio quality, award-winning, 8K resolution, HDR

负向提示

描述你不希望出现的内容，以改善生成结果：

"... avoid blurry, no watermarks, no text overlays, avoid distorted faces"

示例

照片级真实感风景

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -d '{
    "messages": [{"role": "user", "content": "A misty fjord in Norway at sunrise, mirror-like water reflecting snow-capped mountains, a small red fishing cabin on the shore, photorealistic, 8K, golden hour lighting, shot on Hasselblad"}],
    "stream": false
  }'

数字插画

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -d '{
    "messages": [{"role": "user", "content": "A cozy bookstore interior with a cat sleeping on a stack of old books, warm lamp light, plants hanging from the ceiling, digital illustration, Studio Ghibli style, soft colors, whimsical atmosphere"}],
    "stream": false
  }'

产品摄影

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -d '{
    "messages": [{"role": "user", "content": "A sleek wireless earbud case on a marble surface, soft studio lighting, product photography, shallow depth of field, clean minimalist background, commercial quality, 1:1 square format"}],
    "stream": false
  }'

概念艺术

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -d '{
    "messages": [{"role": "user", "content": "A floating cyberpunk city above the clouds, neon lights reflecting off chrome towers, flying vehicles streaming between buildings, concept art, matte painting, epic scale, dramatic perspective, 16:9 cinematic"}],
    "stream": false
  }'

图像中的文字（海报）

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -d '{
    "messages": [{"role": "user", "content": "A vintage concert poster for a band called SOLAR DRIFT, playing at The Echo Lounge, June 15th. Retro psychedelic style with swirling colors, 60s typography, orange and purple palette, portrait format"}],
    "stream": false
  }'

批量生成工作流

# Generate multiple style variations of the same subject
PROMPTS=(
  "A medieval castle on a cliff at sunset, oil painting style"
  "A medieval castle on a cliff at sunset, photorealistic, drone shot"
  "A medieval castle on a cliff at sunset, anime style, Makoto Shinkai"
  "A medieval castle on a cliff at sunset, pixel art, 16-bit"
)

for PROMPT in "${PROMPTS[@]}"; do
  curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
    -H "Content-Type: application/json" \
    -H "X-API-Key: $EACHLABS_API_KEY" \
    -d "{
      \"messages\": [{\"role\": \"user\", \"content\": \"$PROMPT\"}],
      \"stream\": false
    }"
  echo "---"
done

常见问题

• 提示词过于模糊会产生泛泛的结果。请具体描述主体、风格和氛围。
• 单个提示词中包含过多主体会造成混淆。每张图像聚焦于一个主要主体。
• 文字渲染在大多数模型中都不太可靠。对于文字较多的图像，请使用 flux-kontext-pro。
• 手和手指可能会变形。如果手是焦点，请仔细裁剪或设置手部姿势。
• 相互矛盾的风格（例如“照片级真实感水彩”）会让模型困惑。请选择一种风格。
• 提示词内容过多（超过 50 个词）通常会降低品质。建议将描述控制在 15–40 个重点词语。

相关技能

• 图像编辑 — 使用自然语言编辑和修改生成的图像
• 图像放大 — 提升生成图像的分辨率
• 背景移除 — 移除生成图像的背景
• 视频生成 — 将图像制作成动画视频

Documentation

• each::sense Overview
• each::labs API