generateContent 格式生成文本或图片。模型名放在 URL 路径中,请求体内不再填写 model。文本位于 candidates[].content.parts[].text,生成图片位于 candidates[].content.parts[].inlineData.data,其值为 Base64。/v1/models 确认模型名。Gemini 图片模型可以同时返回说明文本和图片。{
"contents": [
{
"role": "user",
"parts": [
{
"text": "解释 RAG 与微调的主要区别。"
}
]
}
],
"generationConfig": {
"temperature": 0.7,
"maxOutputTokens": 1200,
"thinkingConfig": {
"thinkingBudget": 512,
"includeThoughts": false
}
}
}curl --location '/v1beta/models/gemini-3.1-flash-image-preview:generateContent' \
--header 'Content-Type: application/json' \
--data '{
"contents": [
{
"role": "user",
"parts": [
{
"text": "解释 RAG 与微调的主要区别。"
}
]
}
],
"generationConfig": {
"temperature": 0.7,
"maxOutputTokens": 1200,
"thinkingConfig": {
"thinkingBudget": 512,
"includeThoughts": false
}
}
}'inlineData.data,需按 inlineData.mimeType 解码保存。{
"candidates": [
{
"content": {
"role": "model",
"parts": [
{
"text": "RAG 在推理时检索外部资料,微调则改变模型参数。"
}
]
},
"finishReason": "STOP",
"index": 0
}
],
"usageMetadata": {
"promptTokenCount": 18,
"candidatesTokenCount": 42,
"totalTokenCount": 60
}
}