Nano Banana API 不出图：按返回字段逐项排查

# generateContent
curl -s -X POST \
  "https://generativelanguage.googleapis.com/v1/models/gemini-3.1-flash-image:generateContent" \
  -H "x-goog-api-key: $GEMINI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [{"parts": [{"text": "生成一张图片：雨后的老街，暖色路灯，电影感构图"}]}],
    "generationConfig": {"responseModalities": ["TEXT", "IMAGE"]}
  }' \
  -o resp.json -w "HTTP %{http_code}\n"

# Interactions API
curl -s -X POST "https://generativelanguage.googleapis.com/v1beta/interactions" \
  -H "x-goog-api-key: $GEMINI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemini-3.1-flash-image",
    "input": [{"type": "text", "text": "生成一张图片：雨后的老街，暖色路灯，电影感构图"}]
  }' \
  -o interaction.json -w "HTTP %{http_code}\n"

#!/usr/bin/env python3
"""检查一次 Nano Banana 调用的原始 JSON：有没有最终图片、属于哪种“没图”、下一步做什么。
用法：python3 nb_inspect.py resp.json [--save DIR]   只用标准库；退出码 0=有图 1=没图 2=无法识别"""
import base64, json, pathlib, sys

PRICE = {  # 美元 / 100 万 token：输入、文字与思考输出、图片输出（Gemini API 付费层 Standard）
    "gemini-3.1-flash-image": (0.50, 3.0, 60.0),
    "gemini-3-pro-image": (2.0, 12.0, 120.0),
}
POLICY = {"PROHIBITED_CONTENT", "BLOCKLIST", "SPII", "IMAGE_SAFETY", "IMAGE_PROHIBITED_CONTENT"}
ACTION = {
    "prompt_blocked": "提示词在生成前被拦截：看 blockReason，改写内容后再发，不要原样重试",
    "policy_block": "输出被图片安全或政策规则拦截：换题材或描述方式；原样重试没用，四个可调安全类别也不包括它",
    "safety_filter": "可调安全过滤器拦截（SAFETY）：看 safetyRatings；若是你自己设的阈值，放宽它或改写内容",
    "no_image": "模型该出图却没出（NO_IMAGE）：把“生成一张图片”写明，确认请求了图片模态，可再试一次",
    "text_only": "正常结束但只有文字：明确要求出图，并在请求里指定图片输出",
    "recitation": "输出与已有素材过于相似：去掉对具体作品、角色的复刻要求，描述写得更原创",
    "other": "原因未说明（OTHER / IMAGE_OTHER）：保存原始响应，可重试一次；重复出现再改提示词",
    "max_tokens": "输出不完整，常见原因是触到输出上限：去掉或调高 maxOutputTokens / max_output_tokens",
    "thought_signature": "多轮编辑缺少 thought signature：把上一轮响应完整放回历史，或改用 SDK 的 chat",
    "not_finished": "interaction 还没结束：稍后按 id 取回结果再判断",
    "failed": "interaction 失败：保存完整 JSON，查看 steps 与错误信息，不要盲目循环重试",
    "empty": "没有候选也没有拦截原因：保存 responseId 与原始响应，重试一次，重复出现就改提示词",
    "http_error": "请求本身报错（不是“没图”）：429 看额度，其他先核对模型 ID 与参数",
    "ok": "拿到了最终图片；如果你的程序仍说没图，问题在解析代码",
    "unknown": "未知的结束原因：保留原始值，按文档最新枚举核对",
}

def g(d, *keys):
    """同时兼容 REST 的 camelCase 和 SDK 导出的 snake_case。"""
    for k in keys:
        if isinstance(d, dict) and d.get(k) is not None:
            return d[k]
    return None

def by_modality(items):
    return {str(g(i, "modality")).upper(): g(i, "tokenCount", "token_count", "tokens") or 0
            for i in items or []}

def inspect_generate_content(r):
    finals, thoughts, texts, reasons = [], 0, 0, []
    for c in g(r, "candidates") or []:
        reasons.append(g(c, "finishReason", "finish_reason"))
        for p in g(g(c, "content") or {}, "parts") or []:
            blob = g(p, "inlineData", "inline_data")
            is_img = blob and str(g(blob, "mimeType", "mime_type")).startswith("image/")
            if is_img and g(p, "thought"):
                thoughts += 1
            elif is_img:
                finals.append(blob)
            elif g(p, "text") and not g(p, "thought"):
                texts += 1
    fb = g(r, "promptFeedback", "prompt_feedback") or {}
    block = g(fb, "blockReason", "block_reason")
    reason = next((x for x in reasons if x and x != "STOP"), reasons[0] if reasons else None)
    if finals:
        shape = "ok"
    elif block:
        shape = "prompt_blocked"
    elif not reasons:
        shape = "empty"
    elif reason in POLICY:
        shape = "policy_block"
    else:
        shape = {"STOP": "text_only" if texts else "empty", "SAFETY": "safety_filter", "NO_IMAGE": "no_image",
                 "RECITATION": "recitation", "IMAGE_RECITATION": "recitation",
                 "OTHER": "other", "IMAGE_OTHER": "other", "MAX_TOKENS": "max_tokens",
                 "MISSING_THOUGHT_SIGNATURE": "thought_signature"}.get(reason, "unknown")
    u = g(r, "usageMetadata", "usage_metadata") or {}
    out = by_modality(g(u, "candidatesTokensDetails", "candidates_tokens_details"))
    total_out = g(u, "candidatesTokenCount", "candidates_token_count") or 0
    usage = {"input": g(u, "promptTokenCount", "prompt_token_count") or 0,
             "thought": g(u, "thoughtsTokenCount", "thoughts_token_count") or 0,
             "image_out": out.get("IMAGE", 0)}
    usage["text_out"] = total_out - usage["image_out"]
    detail = {"finish_reason": reason, "block_reason": block, "text_parts": texts}
    return shape, finals, thoughts, usage, detail, g(r, "modelVersion", "model_version")

def inspect_interaction(r):
    finals, thoughts, texts = [], 0, 0
    for s in g(r, "steps") or []:
        blocks = g(s, "content") if g(s, "type") == "model_output" else g(s, "summary")
        for b in blocks or []:
            if g(b, "type") == "image" and g(s, "type") == "model_output":
                finals.append(b)
            elif g(b, "type") == "image":
                thoughts += 1
            elif g(b, "type") == "text" and g(s, "type") == "model_output":
                texts += 1
    status = g(r, "status")
    if finals:
        shape = "ok"
    elif status in ("in_progress", "queued", "requires_action"):
        shape = "not_finished"
    elif status in ("failed", "cancelled"):
        shape = "failed"
    elif status == "incomplete":
        shape = "max_tokens"
    elif status == "completed":
        shape = "text_only" if texts else "empty"
    else:
        shape = "unknown"
    u = g(r, "usage") or {}
    out = by_modality(g(u, "output_tokens_by_modality"))
    usage = {"input": g(u, "total_input_tokens") or 0, "thought": g(u, "total_thought_tokens") or 0,
             "image_out": out.get("IMAGE", 0), "text_out": out.get("TEXT", 0)}
    return shape, finals, thoughts, usage, {"status": status, "text_blocks": texts}, g(r, "model")

def estimate(model, usage):
    key = next((k for k in PRICE if model and str(model).split("/")[-1].startswith(k)), None)
    if not key:
        return None
    p_in, p_text, p_img = PRICE[key]
    usd = (usage["input"] * p_in + (usage["thought"] + usage["text_out"]) * p_text
           + usage["image_out"] * p_img) / 1e6
    return round(usd, 6)

def main(argv):
    if len(argv) < 2:
        print(__doc__)
        return 2
    r = json.loads(pathlib.Path(argv[1]).read_text(encoding="utf-8"))
    if isinstance(g(r, "error"), dict):
        e = r["error"]
        print(json.dumps({"api": "error", "shape": "http_error", "code": e.get("code"),
                          "status": e.get("status"), "next": ACTION["http_error"]},
                         ensure_ascii=False, indent=2))
        return 1
    if any(g(r, k) is not None for k in ("candidates", "promptFeedback", "prompt_feedback",
                                         "usageMetadata", "usage_metadata", "modelVersion")):
        api, res = "generateContent", inspect_generate_content(r)
    elif g(r, "steps") is not None or g(r, "status"):
        api, res = "interactions", inspect_interaction(r)
    else:
        print("无法识别：既不是 generateContent 响应，也不是 interaction")
        return 2
    shape, finals, thoughts, usage, detail, model = res
    if "--save" in argv and finals:
        d = pathlib.Path(argv[argv.index("--save") + 1])
        d.mkdir(parents=True, exist_ok=True)
        for i, blob in enumerate(finals, 1):
            ext = str(g(blob, "mimeType", "mime_type") or "image/png").split("/")[-1]
            (d / f"image_{i}.{ext}").write_bytes(base64.b64decode(g(blob, "data")))
    print(json.dumps({"api": api, "model": model, "shape": shape, "final_images": len(finals),
                      "thought_images": thoughts, **detail, "usage_tokens": usage,
                      "estimated_usd": estimate(model, usage), "next": ACTION[shape]},
                     ensure_ascii=False, indent=2))
    return 0 if finals else 1

if __name__ == "__main__":
    sys.exit(main(sys.argv))

python3 nb_inspect.py resp.json                    # 只判断
python3 nb_inspect.py interaction.json --save out  # 有最终图片时存到 out/ 目录

{
  "api": "generateContent",
  "model": "gemini-3.1-flash-image",
  "shape": "policy_block",
  "final_images": 0,
  "thought_images": 0,
  "finish_reason": "IMAGE_SAFETY",
  "block_reason": null,
  "text_parts": 0,
  "usage_tokens": {
    "input": 12,
    "thought": 180,
    "image_out": 0,
    "text_out": 0
  },
  "estimated_usd": 0.000546,
  "next": "输出被图片安全或政策规则拦截：换题材或描述方式；原样重试没用，四个可调安全类别也不包括它"
}