zhoujie
0c91c00dcf
- 在系统提示词中新增【反AI检测规则】章节，包含句子长度、逻辑跳跃、标点符号等8项防检测措施 - 为所有SD模型提示词指南添加通用反AI检测技巧，强调真实手机拍摄风格 - 深度重构 `_humanize_content` 方法，新增8层真人化处理：替换书面表达、打散句子长度、随机添加口语元素、模拟手机打字标点习惯 - 增强 `_humanize` 方法，去除更多AI前缀，随机化标点，限制表情符号堆叠 - 在 `sd_service.py` 新增 `anti_detect_postprocess` 图片后处理管线，包含元数据剥离、随机裁剪、色彩微扰、不均匀噪声、JPEG压缩回环等7步处理 - 所有图片生成后自动经过反检测处理，输出格式统一为JPEG以模拟真实手机照片 - 更新 `main.py` 中的图片保存逻辑，统一使用JPEG格式并确保RGB模式转换
639 lines
26 KiB
Python
639 lines
26 KiB
Python
"""
|
||
Stable Diffusion 服务模块
|
||
封装对 SD WebUI API 的调用，支持 txt2img 和 img2img，支持 ReActor 换脸
|
||
含图片反 AI 检测后处理管线
|
||
"""
|
||
import requests
|
||
import base64
|
||
import io
|
||
import logging
|
||
import os
|
||
import random
|
||
import math
|
||
import struct
|
||
import zlib
|
||
from PIL import Image, ImageFilter, ImageEnhance
|
||
|
||
logger = logging.getLogger(__name__)
|
||
|
||
SD_TIMEOUT = 1800 # 图片生成可能需要较长时间
|
||
|
||
# 头像文件默认保存路径
|
||
FACE_IMAGE_PATH = os.path.join(os.path.dirname(__file__), "my_face.png")
|
||
|
||
# ==================== 多模型配置系统 ====================
|
||
# 每个模型的最优参数、prompt 增强词、负面提示词、三档预设
|
||
|
||
SD_MODEL_PROFILES = {
|
||
# ---- majicmixRealistic: 东亚网红感，朋友圈自拍/美妆/穿搭 (SD 1.5) ----
|
||
"majicmixRealistic": {
|
||
"display_name": "majicmixRealistic ⭐⭐⭐⭐⭐",
|
||
"description": "东亚网红感 | 朋友圈自拍、美妆、穿搭",
|
||
"arch": "sd15", # SD 1.5 架构
|
||
# 自动追加到 prompt 前面的增强词
|
||
"prompt_prefix": (
|
||
"(best quality:1.4), (masterpiece:1.4), (ultra detailed:1.3), "
|
||
"(photorealistic:1.4), (realistic:1.3), raw photo, "
|
||
"(asian girl:1.3), (chinese:1.2), (east asian features:1.2), "
|
||
"(delicate facial features:1.2), (fair skin:1.1), (natural skin texture:1.2), "
|
||
"(soft lighting:1.1), (natural makeup:1.1), "
|
||
),
|
||
# 自动追加到 prompt 后面的补充词
|
||
"prompt_suffix": (
|
||
", film grain, shallow depth of field, "
|
||
"instagram aesthetic, xiaohongshu style, phone camera feel"
|
||
),
|
||
"negative_prompt": (
|
||
"(nsfw:1.5), (nudity:1.5), (worst quality:2), (low quality:2), (normal quality:2), "
|
||
"lowres, bad anatomy, bad hands, text, error, missing fingers, "
|
||
"extra digit, fewer digits, cropped, jpeg artifacts, signature, watermark, "
|
||
"blurry, deformed, mutated, disfigured, ugly, duplicate, "
|
||
"poorly drawn face, poorly drawn hands, extra limbs, fused fingers, "
|
||
"too many fingers, long neck, out of frame, "
|
||
"western face, european face, caucasian, deep-set eyes, high nose bridge, "
|
||
"blonde hair, red hair, blue eyes, green eyes, freckles, thick body hair, "
|
||
"painting, cartoon, anime, sketch, illustration, 3d render"
|
||
),
|
||
"presets": {
|
||
"快速 (约30秒)": {
|
||
"steps": 20,
|
||
"cfg_scale": 7.0,
|
||
"width": 512,
|
||
"height": 768,
|
||
"sampler_name": "Euler a",
|
||
"scheduler": "Normal",
|
||
"batch_size": 2,
|
||
},
|
||
"标准 (约1分钟)": {
|
||
"steps": 30,
|
||
"cfg_scale": 7.0,
|
||
"width": 512,
|
||
"height": 768,
|
||
"sampler_name": "DPM++ 2M",
|
||
"scheduler": "Karras",
|
||
"batch_size": 2,
|
||
},
|
||
"精细 (约2-3分钟)": {
|
||
"steps": 40,
|
||
"cfg_scale": 7.5,
|
||
"width": 576,
|
||
"height": 864,
|
||
"sampler_name": "DPM++ SDE",
|
||
"scheduler": "Karras",
|
||
"batch_size": 2,
|
||
},
|
||
},
|
||
},
|
||
|
||
# ---- Realistic Vision: 写实摄影感，纪实摄影/街拍/真实质感 (SD 1.5) ----
|
||
"realisticVision": {
|
||
"display_name": "Realistic Vision ⭐⭐⭐⭐",
|
||
"description": "写实摄影感 | 纪实摄影、街拍、真实质感",
|
||
"arch": "sd15",
|
||
"prompt_prefix": (
|
||
"RAW photo, (best quality:1.4), (masterpiece:1.3), (realistic:1.4), "
|
||
"(photorealistic:1.4), 8k uhd, DSLR, high quality, "
|
||
"(asian:1.2), (chinese girl:1.2), (east asian features:1.1), "
|
||
"(natural skin:1.2), (skin pores:1.1), (detailed skin texture:1.2), "
|
||
),
|
||
"prompt_suffix": (
|
||
", shot on Canon EOS R5, 85mm lens, f/1.8, "
|
||
"natural lighting, documentary style, street photography, "
|
||
"film color grading, depth of field"
|
||
),
|
||
"negative_prompt": (
|
||
"(nsfw:1.5), (nudity:1.5), (worst quality:2), (low quality:2), (normal quality:2), "
|
||
"lowres, bad anatomy, bad hands, text, error, missing fingers, "
|
||
"extra digit, fewer digits, cropped, jpeg artifacts, signature, watermark, "
|
||
"blurry, deformed, mutated, disfigured, ugly, duplicate, "
|
||
"poorly drawn face, extra limbs, fused fingers, long neck, "
|
||
"western face, european face, caucasian, deep-set eyes, "
|
||
"blonde hair, blue eyes, green eyes, freckles, "
|
||
"painting, cartoon, anime, sketch, illustration, 3d render, "
|
||
"over-sharpened, over-saturated, plastic skin, airbrushed, "
|
||
"smooth skin, doll-like, HDR, overprocessed"
|
||
),
|
||
"presets": {
|
||
"快速 (约30秒)": {
|
||
"steps": 20,
|
||
"cfg_scale": 7.0,
|
||
"width": 512,
|
||
"height": 768,
|
||
"sampler_name": "Euler a",
|
||
"scheduler": "Normal",
|
||
"batch_size": 2,
|
||
},
|
||
"标准 (约1分钟)": {
|
||
"steps": 28,
|
||
"cfg_scale": 7.0,
|
||
"width": 512,
|
||
"height": 768,
|
||
"sampler_name": "DPM++ 2M",
|
||
"scheduler": "Karras",
|
||
"batch_size": 2,
|
||
},
|
||
"精细 (约2-3分钟)": {
|
||
"steps": 40,
|
||
"cfg_scale": 7.5,
|
||
"width": 576,
|
||
"height": 864,
|
||
"sampler_name": "DPM++ SDE",
|
||
"scheduler": "Karras",
|
||
"batch_size": 2,
|
||
},
|
||
},
|
||
},
|
||
|
||
# ---- Juggernaut XL: 电影大片感，高画质/商业摄影/复杂背景 (SDXL) ----
|
||
"juggernautXL": {
|
||
"display_name": "Juggernaut XL ⭐⭐⭐⭐",
|
||
"description": "电影大片感 | 高画质、商业摄影、复杂背景",
|
||
"arch": "sdxl", # SDXL 架构
|
||
"prompt_prefix": (
|
||
"masterpiece, best quality, ultra detailed, 8k uhd, high resolution, "
|
||
"photorealistic, cinematic lighting, cinematic composition, "
|
||
"asian girl, chinese, east asian features, black hair, dark brown eyes, "
|
||
"delicate facial features, fair skin, slim figure, "
|
||
),
|
||
"prompt_suffix": (
|
||
", cinematic color grading, anamorphic lens, bokeh, "
|
||
"volumetric lighting, ray tracing, global illumination, "
|
||
"commercial photography, editorial style, vogue aesthetic"
|
||
),
|
||
"negative_prompt": (
|
||
"nsfw, nudity, lowres, bad anatomy, bad hands, text, error, missing fingers, "
|
||
"extra digit, fewer digits, cropped, worst quality, low quality, normal quality, "
|
||
"jpeg artifacts, signature, watermark, blurry, deformed, mutated, disfigured, "
|
||
"ugly, duplicate, morbid, mutilated, poorly drawn face, poorly drawn hands, "
|
||
"extra limbs, fused fingers, too many fingers, long neck, username, "
|
||
"out of frame, distorted, oversaturated, underexposed, overexposed, "
|
||
"western face, european face, caucasian, deep-set eyes, high nose bridge, "
|
||
"blonde hair, red hair, blue eyes, green eyes, freckles, thick body hair"
|
||
),
|
||
"presets": {
|
||
"快速 (约30秒)": {
|
||
"steps": 12,
|
||
"cfg_scale": 5.0,
|
||
"width": 768,
|
||
"height": 1024,
|
||
"sampler_name": "Euler a",
|
||
"scheduler": "Normal",
|
||
"batch_size": 2,
|
||
},
|
||
"标准 (约1分钟)": {
|
||
"steps": 20,
|
||
"cfg_scale": 5.5,
|
||
"width": 832,
|
||
"height": 1216,
|
||
"sampler_name": "DPM++ 2M",
|
||
"scheduler": "Karras",
|
||
"batch_size": 2,
|
||
},
|
||
"精细 (约2-3分钟)": {
|
||
"steps": 35,
|
||
"cfg_scale": 6.0,
|
||
"width": 832,
|
||
"height": 1216,
|
||
"sampler_name": "DPM++ 2M SDE",
|
||
"scheduler": "Karras",
|
||
"batch_size": 2,
|
||
},
|
||
},
|
||
},
|
||
}
|
||
|
||
# 默认配置 profile key
|
||
DEFAULT_MODEL_PROFILE = "juggernautXL"
|
||
|
||
|
||
def detect_model_profile(model_name: str) -> str:
|
||
"""根据 SD 模型名称自动识别对应的 profile key"""
|
||
name_lower = model_name.lower() if model_name else ""
|
||
if "majicmix" in name_lower or "majic" in name_lower:
|
||
return "majicmixRealistic"
|
||
elif "realistic" in name_lower and "vision" in name_lower:
|
||
return "realisticVision"
|
||
elif "rv" in name_lower and ("v5" in name_lower or "v6" in name_lower or "v4" in name_lower):
|
||
return "realisticVision" # RV v5.1 等简写
|
||
elif "juggernaut" in name_lower or "jugger" in name_lower:
|
||
return "juggernautXL"
|
||
# 根据架构猜测
|
||
elif "xl" in name_lower or "sdxl" in name_lower:
|
||
return "juggernautXL" # SDXL 架构默认用 Juggernaut 参数
|
||
else:
|
||
return DEFAULT_MODEL_PROFILE # 无法识别时默认
|
||
|
||
|
||
def get_model_profile(model_name: str = None) -> dict:
|
||
"""获取模型配置 profile"""
|
||
key = detect_model_profile(model_name) if model_name else DEFAULT_MODEL_PROFILE
|
||
return SD_MODEL_PROFILES.get(key, SD_MODEL_PROFILES[DEFAULT_MODEL_PROFILE])
|
||
|
||
|
||
def get_model_profile_info(model_name: str = None) -> str:
|
||
"""获取当前模型的显示信息 (Markdown 格式)"""
|
||
profile = get_model_profile(model_name)
|
||
key = detect_model_profile(model_name) if model_name else DEFAULT_MODEL_PROFILE
|
||
is_default = key == DEFAULT_MODEL_PROFILE and model_name and detect_model_profile(model_name) == DEFAULT_MODEL_PROFILE
|
||
# 如果检测结果是默认回退的, 说明是未知模型
|
||
actual_key = detect_model_profile(model_name) if model_name else None
|
||
presets = profile["presets"]
|
||
first_preset = list(presets.values())[0]
|
||
res = f"{first_preset.get('width', '?')}×{first_preset.get('height', '?')}"
|
||
lines = [
|
||
f"**🎨 {profile['display_name']}** | `{profile['arch'].upper()}` | {res}",
|
||
f"> {profile['description']}",
|
||
]
|
||
if model_name and not any(k in (model_name or "").lower() for k in ["majicmix", "realistic", "juggernaut"]):
|
||
lines.append(f"> ⚠️ 未识别的模型，使用默认档案 ({profile['display_name']})")
|
||
return "\n".join(lines)
|
||
|
||
|
||
# ==================== 兼容旧接口 ====================
|
||
# 默认预设和反向提示词 (使用 Juggernaut XL 作为默认)
|
||
|
||
SD_PRESETS = SD_MODEL_PROFILES[DEFAULT_MODEL_PROFILE]["presets"]
|
||
SD_PRESET_NAMES = list(SD_PRESETS.keys())
|
||
|
||
|
||
def get_sd_preset(name: str, model_name: str = None) -> dict:
|
||
"""获取生成预设参数，自动适配模型"""
|
||
profile = get_model_profile(model_name)
|
||
presets = profile.get("presets", SD_PRESETS)
|
||
return presets.get(name, presets.get("标准 (约1分钟)", list(presets.values())[0]))
|
||
|
||
|
||
# 默认反向提示词（Juggernaut XL）
|
||
DEFAULT_NEGATIVE = SD_MODEL_PROFILES[DEFAULT_MODEL_PROFILE]["negative_prompt"]
|
||
|
||
|
||
# ==================== 图片反 AI 检测管线 ====================
|
||
|
||
def anti_detect_postprocess(img: Image.Image) -> Image.Image:
|
||
"""对 AI 生成的图片进行后处理，模拟手机拍摄/加工特征，降低 AI 检测率
|
||
|
||
处理流程:
|
||
1. 剥离所有元数据 (EXIF/SD参数/PNG chunks)
|
||
2. 微小随机裁剪 (模拟手机截图不完美)
|
||
3. 微小旋转+校正 (破坏像素完美对齐)
|
||
4. 色彩微扰 (模拟手机屏幕色差)
|
||
5. 不均匀高斯噪声 (模拟传感器噪声)
|
||
6. 微小锐化 (模拟手机锐化算法)
|
||
7. JPEG 压缩回环 (最关键: 引入真实压缩伪影)
|
||
"""
|
||
if img.mode != "RGB":
|
||
img = img.convert("RGB")
|
||
|
||
w, h = img.size
|
||
|
||
# Step 1: 微小随机裁剪 (1-3 像素, 破坏边界对齐)
|
||
crop_l = random.randint(0, 3)
|
||
crop_t = random.randint(0, 3)
|
||
crop_r = random.randint(0, 3)
|
||
crop_b = random.randint(0, 3)
|
||
if crop_l + crop_r < w and crop_t + crop_b < h:
|
||
img = img.crop((crop_l, crop_t, w - crop_r, h - crop_b))
|
||
|
||
# Step 2: 极微旋转 (0.1°-0.5°, 破坏完美像素排列)
|
||
if random.random() < 0.6:
|
||
angle = random.uniform(-0.5, 0.5)
|
||
img = img.rotate(angle, resample=Image.BICUBIC, expand=False,
|
||
fillcolor=(
|
||
random.randint(240, 255),
|
||
random.randint(240, 255),
|
||
random.randint(240, 255),
|
||
))
|
||
|
||
# Step 3: 色彩微扰 (模拟手机屏幕/相机色差)
|
||
# 亮度微调
|
||
brightness_factor = random.uniform(0.97, 1.03)
|
||
img = ImageEnhance.Brightness(img).enhance(brightness_factor)
|
||
# 对比度微调
|
||
contrast_factor = random.uniform(0.97, 1.03)
|
||
img = ImageEnhance.Contrast(img).enhance(contrast_factor)
|
||
# 饱和度微调
|
||
saturation_factor = random.uniform(0.96, 1.04)
|
||
img = ImageEnhance.Color(img).enhance(saturation_factor)
|
||
|
||
# Step 4: 不均匀传感器噪声 (比均匀噪声更像真实相机)
|
||
try:
|
||
import numpy as np
|
||
arr = np.array(img, dtype=np.float32)
|
||
# 生成不均匀噪声: 中心弱边缘强 (模拟暗角)
|
||
h_arr, w_arr = arr.shape[:2]
|
||
y_grid, x_grid = np.mgrid[0:h_arr, 0:w_arr]
|
||
center_y, center_x = h_arr / 2, w_arr / 2
|
||
dist = np.sqrt((y_grid - center_y) ** 2 + (x_grid - center_x) ** 2)
|
||
max_dist = np.sqrt(center_y ** 2 + center_x ** 2)
|
||
# 噪声强度: 中心 1.0, 边缘 2.5
|
||
noise_strength = 1.0 + 1.5 * (dist / max_dist)
|
||
noise_strength = noise_strength[:, :, np.newaxis]
|
||
# 高斯噪声
|
||
noise = np.random.normal(0, random.uniform(1.5, 3.0), arr.shape) * noise_strength
|
||
arr = np.clip(arr + noise, 0, 255).astype(np.uint8)
|
||
img = Image.fromarray(arr)
|
||
except ImportError:
|
||
# numpy 不可用时用 PIL 的简单模糊代替
|
||
pass
|
||
|
||
# Step 5: 轻微锐化 (模拟手机后处理)
|
||
if random.random() < 0.5:
|
||
img = img.filter(ImageFilter.SHARPEN)
|
||
# 再做一次轻微模糊中和, 避免过度锐化
|
||
img = img.filter(ImageFilter.GaussianBlur(radius=0.3))
|
||
|
||
# Step 6: JPEG 压缩回环 (最关键! 引入真实压缩伪影)
|
||
# 模拟: 手机保存 → 社交平台压缩 → 重新上传
|
||
quality = random.randint(85, 93) # 质量略低于完美, 像手机存储
|
||
buf = io.BytesIO()
|
||
img.save(buf, format="JPEG", quality=quality, subsampling=0)
|
||
buf.seek(0)
|
||
img = Image.open(buf).copy() # 重新加载, 已包含 JPEG 伪影
|
||
|
||
# Step 7: resize 回原始尺寸附近 (模拟平台缩放)
|
||
# 微小缩放 ±2%
|
||
scale = random.uniform(0.98, 1.02)
|
||
new_w = int(img.width * scale)
|
||
new_h = int(img.height * scale)
|
||
if new_w > 100 and new_h > 100:
|
||
img = img.resize((new_w, new_h), Image.LANCZOS)
|
||
|
||
logger.info("🛡️ 图片反检测后处理完成: crop=%dx%d→%dx%d, jpeg_q=%d, scale=%.2f",
|
||
w, h, img.width, img.height, quality, scale)
|
||
return img
|
||
|
||
|
||
def strip_metadata(img: Image.Image) -> Image.Image:
|
||
"""彻底剥离图片所有元数据 (EXIF, SD参数, PNG text chunks)"""
|
||
clean = Image.new(img.mode, img.size)
|
||
clean.putdata(list(img.getdata()))
|
||
return clean
|
||
|
||
|
||
class SDService:
|
||
"""Stable Diffusion WebUI API 封装"""
|
||
|
||
def __init__(self, sd_url: str = "http://127.0.0.1:7860"):
|
||
self.sd_url = sd_url.rstrip("/")
|
||
|
||
# ---------- 工具方法 ----------
|
||
|
||
@staticmethod
|
||
def _image_to_base64(img: Image.Image) -> str:
|
||
"""PIL Image → base64 字符串"""
|
||
buf = io.BytesIO()
|
||
img.save(buf, format="PNG")
|
||
return base64.b64encode(buf.getvalue()).decode("utf-8")
|
||
|
||
@staticmethod
|
||
def load_face_image(path: str = None) -> Image.Image | None:
|
||
"""加载头像图片，不存在则返回 None"""
|
||
path = path or FACE_IMAGE_PATH
|
||
if path and os.path.isfile(path):
|
||
try:
|
||
return Image.open(path).convert("RGB")
|
||
except Exception as e:
|
||
logger.warning("头像加载失败: %s", e)
|
||
return None
|
||
|
||
@staticmethod
|
||
def save_face_image(img: Image.Image, path: str = None) -> str:
|
||
"""保存头像图片，返回保存路径"""
|
||
path = path or FACE_IMAGE_PATH
|
||
img = img.convert("RGB")
|
||
img.save(path, format="PNG")
|
||
logger.info("头像已保存: %s", path)
|
||
return path
|
||
|
||
def _build_reactor_args(self, face_image: Image.Image) -> dict:
|
||
"""构建 ReActor 换脸参数（alwayson_scripts 格式）
|
||
|
||
参数索引对照 (reactor script-info):
|
||
0: source_image (base64) 1: enable 2: source_faces
|
||
3: target_faces 4: model 5: restore_face
|
||
6: restore_visibility 7: restore_first 8: upscaler
|
||
9: scale 10: upscaler_vis 11: swap_in_source
|
||
12: swap_in_generated 13: log_level 14: gender_source
|
||
15: gender_target 16: save_original 17: codeformer_weight
|
||
18: source_hash_check 19: target_hash_check 20: exec_provider
|
||
21: face_mask_correction 22: select_source 23: face_model
|
||
24: source_folder 25: multiple_sources 26: random_image
|
||
27: force_upscale 28: threshold 29: max_faces
|
||
30: tab_single
|
||
"""
|
||
face_b64 = self._image_to_base64(face_image)
|
||
return {
|
||
"reactor": {
|
||
"args": [
|
||
face_b64, # 0: source image (base64)
|
||
True, # 1: enable ReActor
|
||
"0", # 2: source face index
|
||
"0", # 3: target face index
|
||
"inswapper_128.onnx", # 4: swap model
|
||
"CodeFormer", # 5: restore face method
|
||
1, # 6: restore face visibility
|
||
True, # 7: restore face first, then upscale
|
||
"None", # 8: upscaler
|
||
1, # 9: scale
|
||
1, # 10: upscaler visibility
|
||
False, # 11: swap in source
|
||
True, # 12: swap in generated
|
||
1, # 13: console log level (0=min, 1=med, 2=max)
|
||
0, # 14: gender detection source (0=No)
|
||
0, # 15: gender detection target (0=No)
|
||
False, # 16: save original
|
||
0.8, # 17: CodeFormer weight (0=max effect, 1=min)
|
||
False, # 18: source hash check
|
||
False, # 19: target hash check
|
||
"CUDA", # 20: execution provider
|
||
True, # 21: face mask correction
|
||
0, # 22: select source (0=Image, 1=FaceModel, 2=Folder)
|
||
"", # 23: face model filename (when #22=1)
|
||
"", # 24: source folder path (when #22=2)
|
||
None, # 25: skip for API
|
||
False, # 26: random image
|
||
False, # 27: force upscale
|
||
0.6, # 28: face detection threshold
|
||
2, # 29: max faces to detect (0=unlimited)
|
||
],
|
||
}
|
||
}
|
||
|
||
def has_reactor(self) -> bool:
|
||
"""检查 SD WebUI 是否安装了 ReActor 扩展"""
|
||
try:
|
||
resp = requests.get(f"{self.sd_url}/sdapi/v1/scripts", timeout=5)
|
||
scripts = resp.json()
|
||
all_scripts = scripts.get("txt2img", []) + scripts.get("img2img", [])
|
||
return any("reactor" in s.lower() for s in all_scripts)
|
||
except Exception:
|
||
return False
|
||
|
||
def check_connection(self) -> tuple[bool, str]:
|
||
"""检查 SD 服务是否可用"""
|
||
try:
|
||
resp = requests.get(f"{self.sd_url}/sdapi/v1/sd-models", timeout=5)
|
||
if resp.status_code == 200:
|
||
count = len(resp.json())
|
||
return True, f"SD 已连接，{count} 个模型可用"
|
||
return False, f"SD 返回异常状态: {resp.status_code}"
|
||
except requests.exceptions.ConnectionError:
|
||
return False, "SD WebUI 未启动或端口错误"
|
||
except Exception as e:
|
||
return False, f"SD 连接失败: {e}"
|
||
|
||
def get_models(self) -> list[str]:
|
||
"""获取 SD 模型列表"""
|
||
resp = requests.get(f"{self.sd_url}/sdapi/v1/sd-models", timeout=5)
|
||
resp.raise_for_status()
|
||
return [m["title"] for m in resp.json()]
|
||
|
||
def switch_model(self, model_name: str):
|
||
"""切换 SD 模型"""
|
||
try:
|
||
requests.post(
|
||
f"{self.sd_url}/sdapi/v1/options",
|
||
json={"sd_model_checkpoint": model_name},
|
||
timeout=60,
|
||
)
|
||
except Exception as e:
|
||
logger.warning("模型切换失败: %s", e)
|
||
|
||
def txt2img(
|
||
self,
|
||
prompt: str,
|
||
negative_prompt: str = None,
|
||
model: str = None,
|
||
steps: int = None,
|
||
cfg_scale: float = None,
|
||
width: int = None,
|
||
height: int = None,
|
||
batch_size: int = None,
|
||
seed: int = -1,
|
||
sampler_name: str = None,
|
||
scheduler: str = None,
|
||
face_image: Image.Image = None,
|
||
quality_mode: str = None,
|
||
) -> list[Image.Image]:
|
||
"""文生图（自动适配当前 SD 模型的最优参数）
|
||
|
||
Args:
|
||
model: SD 模型名，自动识别并应用对应配置
|
||
face_image: 头像 PIL Image，传入后自动启用 ReActor 换脸
|
||
quality_mode: 预设模式名
|
||
"""
|
||
if model:
|
||
self.switch_model(model)
|
||
|
||
# 自动识别模型配置
|
||
profile = get_model_profile(model)
|
||
profile_key = detect_model_profile(model)
|
||
logger.info("🎯 SD 模型识别: %s → %s (%s)",
|
||
model or "默认", profile_key, profile["description"])
|
||
|
||
# 加载模型专属预设参数
|
||
preset = get_sd_preset(quality_mode, model) if quality_mode else get_sd_preset("标准 (约1分钟)", model)
|
||
|
||
# 自动增强 prompt: 前缀 + 原始 prompt + 后缀
|
||
enhanced_prompt = profile.get("prompt_prefix", "") + prompt + profile.get("prompt_suffix", "")
|
||
|
||
# 使用模型专属反向提示词
|
||
final_negative = negative_prompt if negative_prompt is not None else profile.get("negative_prompt", DEFAULT_NEGATIVE)
|
||
|
||
payload = {
|
||
"prompt": enhanced_prompt,
|
||
"negative_prompt": final_negative,
|
||
"steps": steps if steps is not None else preset["steps"],
|
||
"cfg_scale": cfg_scale if cfg_scale is not None else preset["cfg_scale"],
|
||
"width": width if width is not None else preset["width"],
|
||
"height": height if height is not None else preset["height"],
|
||
"batch_size": batch_size if batch_size is not None else preset["batch_size"],
|
||
"seed": seed,
|
||
"sampler_name": sampler_name if sampler_name is not None else preset["sampler_name"],
|
||
"scheduler": scheduler if scheduler is not None else preset["scheduler"],
|
||
}
|
||
logger.info("SD 生成参数 [%s]: steps=%s, cfg=%.1f, %dx%d, sampler=%s",
|
||
profile_key, payload['steps'], payload['cfg_scale'],
|
||
payload['width'], payload['height'], payload['sampler_name'])
|
||
|
||
# 如果提供了头像，通过 ReActor 换脸
|
||
if face_image is not None:
|
||
payload["alwayson_scripts"] = self._build_reactor_args(face_image)
|
||
logger.info("🎭 ReActor 换脸已启用")
|
||
|
||
resp = requests.post(
|
||
f"{self.sd_url}/sdapi/v1/txt2img",
|
||
json=payload,
|
||
timeout=SD_TIMEOUT,
|
||
)
|
||
resp.raise_for_status()
|
||
|
||
images = []
|
||
for img_b64 in resp.json().get("images", []):
|
||
img = Image.open(io.BytesIO(base64.b64decode(img_b64)))
|
||
# 反 AI 检测后处理: 剥离元数据 + 模拟手机拍摄特征
|
||
img = anti_detect_postprocess(img)
|
||
images.append(img)
|
||
return images
|
||
|
||
def img2img(
|
||
self,
|
||
init_image: Image.Image,
|
||
prompt: str,
|
||
negative_prompt: str = None,
|
||
denoising_strength: float = 0.5,
|
||
steps: int = 30,
|
||
cfg_scale: float = None,
|
||
sampler_name: str = None,
|
||
scheduler: str = None,
|
||
model: str = None,
|
||
) -> list[Image.Image]:
|
||
"""图生图（自动适配模型参数）"""
|
||
profile = get_model_profile(model)
|
||
preset = get_sd_preset("标准 (约1分钟)", model)
|
||
|
||
# 将 PIL Image 转为 base64
|
||
buf = io.BytesIO()
|
||
init_image.save(buf, format="PNG")
|
||
init_b64 = base64.b64encode(buf.getvalue()).decode("utf-8")
|
||
|
||
enhanced_prompt = profile.get("prompt_prefix", "") + prompt + profile.get("prompt_suffix", "")
|
||
final_negative = negative_prompt if negative_prompt is not None else profile.get("negative_prompt", DEFAULT_NEGATIVE)
|
||
|
||
payload = {
|
||
"init_images": [init_b64],
|
||
"prompt": enhanced_prompt,
|
||
"negative_prompt": final_negative,
|
||
"denoising_strength": denoising_strength,
|
||
"steps": steps,
|
||
"cfg_scale": cfg_scale if cfg_scale is not None else preset["cfg_scale"],
|
||
"width": init_image.width,
|
||
"height": init_image.height,
|
||
"sampler_name": sampler_name if sampler_name is not None else preset["sampler_name"],
|
||
"scheduler": scheduler if scheduler is not None else preset["scheduler"],
|
||
}
|
||
|
||
resp = requests.post(
|
||
f"{self.sd_url}/sdapi/v1/img2img",
|
||
json=payload,
|
||
timeout=SD_TIMEOUT,
|
||
)
|
||
resp.raise_for_status()
|
||
|
||
images = []
|
||
for img_b64 in resp.json().get("images", []):
|
||
img = Image.open(io.BytesIO(base64.b64decode(img_b64)))
|
||
# 反 AI 检测后处理
|
||
img = anti_detect_postprocess(img)
|
||
images.append(img)
|
||
return images
|
||
|
||
def get_lora_models(self) -> list[str]:
|
||
"""获取可用的 LoRA 模型列表"""
|
||
try:
|
||
resp = requests.get(f"{self.sd_url}/sdapi/v1/loras", timeout=5)
|
||
resp.raise_for_status()
|
||
return [lora["name"] for lora in resp.json()]
|
||
except Exception:
|
||
return []
|