update

2026-05-14 13:30:06 +08:00
parent 974946cee4
commit e43171521d
91 changed files with 10485 additions and 3254 deletions
@@ -0,0 +1,93 @@
+from __future__ import annotations
+
+from typing import Optional
+
+from pydantic import BaseModel, Field
+
+
+class ImageInput(BaseModel):
+    image_b64: str = Field(..., description="PNG/JPG 编码后的 base64（不含 data: 前缀）")
+    model_name: Optional[str] = Field(None, description="模型 key（来自 /models）")
+
+
+class DepthRequest(ImageInput):
+    pass
+
+
+class SegmentRequest(ImageInput):
+    pass
+
+
+class SamPromptSegmentRequest(BaseModel):
+    image_b64: str = Field(..., description="裁剪后的 RGB 图 base64（PNG/JPG）")
+    overlay_b64: Optional[str] = Field(
+        None,
+        description="与裁剪同尺寸的标记叠加 PNG base64（可选；当前用于校验尺寸一致）",
+    )
+    point_coords: list[list[float]] = Field(
+        ...,
+        description="裁剪坐标系下的提示点 [[x,y], ...]",
+    )
+    point_labels: list[int] = Field(
+        ...,
+        description="与 point_coords 等长：1=前景，0=背景",
+    )
+    box_xyxy: list[float] = Field(
+        ...,
+        description="裁剪内笔画紧包围盒 [x1,y1,x2,y2]（像素）",
+        min_length=4,
+        max_length=4,
+    )
+    expand_px: int = Field(
+        2,
+        ge=0,
+        le=20,
+        description="对输出 mask 做轻微膨胀的像素半径（0 表示不扩大；建议 1~3）",
+    )
+
+
+class InpaintRequest(ImageInput):
+    prompt: Optional[str] = Field("", description="补全 prompt")
+    strength: float = Field(0.8, ge=0.0, le=1.0)
+    negative_prompt: Optional[str] = Field("", description="负向 prompt")
+    mask_b64: Optional[str] = Field(None, description="mask PNG base64（可选）")
+    max_side: int = Field(1024, ge=128, le=2048)
+
+
+class AnimateRequest(BaseModel):
+    model_name: Optional[str] = Field(None, description="模型 key（来自 /models）")
+    prompt: str = Field(..., description="文本提示词")
+    negative_prompt: Optional[str] = Field("", description="负向提示词")
+    num_inference_steps: int = Field(25, ge=1, le=200)
+    guidance_scale: float = Field(8.0, ge=0.0, le=30.0)
+    width: int = Field(512, ge=128, le=2048)
+    height: int = Field(512, ge=128, le=2048)
+    video_length: int = Field(16, ge=1, le=128)
+    seed: int = Field(-1, description="-1 表示随机种子")
+
+
+class CharacterAnimateRequest(BaseModel):
+    image_b64: str = Field(..., description="透明背景角色 PNG 的 base64（不含 data: 前缀）")
+    model_name: Optional[str] = Field(None, description="模型 key（来自 /models）")
+    prompt: str = Field(..., description="角色动画提示词")
+    negative_prompt: Optional[str] = Field("", description="负向提示词")
+    background_color: str = Field(
+        "#00FF00",
+        description="生成时使用的纯色背景，支持 #RRGGBB 或 R,G,B",
+    )
+    background_tolerance: int = Field(
+        40,
+        ge=0,
+        le=255,
+        description="自动剔除背景的 RGB 色差阈值，越大剔除越激进",
+    )
+    num_inference_steps: int = Field(25, ge=1, le=200)
+    guidance_scale: float = Field(8.0, ge=0.0, le=30.0)
+    video_length: int = Field(16, ge=1, le=128, description="前端请求生成的帧数")
+    seed: int = Field(-1, description="-1 表示随机种子")
+    max_side: int = Field(
+        768,
+        ge=128,
+        le=2048,
+        description="推理分辨率长边上限；会保持输入比例并对齐到 8 的倍数",
+    )