大量修改二期功能清单内容

2026-08-17 18:26:42 +08:00
parent 36e91aab3c
commit d1ecb52125
76 changed files with 4222 additions and 294 deletions
@@ -0,0 +1,77 @@
+"""上传脚本取正文 —— 只认 docx / txt。
+
+docx 用标准库拆(zip + xml),不引 python-docx:镜像少一个依赖,也避免它对损坏文件抛一堆内部异常。
+段落 = <w:p>,文字 = 其中所有 <w:t> 拼接;<w:br>/<w:tab> 补空白,否则整段会粘成一坨。
+"""
+
+import re
+import zipfile
+from io import BytesIO
+from xml.etree import ElementTree
+
+W_NS = "http://schemas.openxmlformats.org/wordprocessingml/2006/main"
+MAX_UPLOAD_BYTES = 5 * 1024 * 1024  # 脚本是纯文字,5MB 足够;更大的基本是误传
+
+
+class ScriptFileError(ValueError):
+    """取正文失败,message 直接给用户看(中文)。"""
+
+
+def _docx_text(raw: bytes) -> str:
+    try:
+        with zipfile.ZipFile(BytesIO(raw)) as bundle:
+            document = bundle.read("word/document.xml")
+    except KeyError:
+        raise ScriptFileError("这个 docx 里没有正文,请确认文件没有损坏")
+    except zipfile.BadZipFile:
+        raise ScriptFileError("这个文件不是有效的 docx,请另存为 .docx 后重试")
+
+    try:
+        root = ElementTree.fromstring(document)
+    except ElementTree.ParseError:
+        raise ScriptFileError("docx 正文解析失败,请另存一份后重试")
+
+    paragraphs = []
+    for node in root.iter(f"{{{W_NS}}}p"):
+        pieces = []
+        for child in node.iter():
+            tag = child.tag
+            if tag == f"{{{W_NS}}}t":
+                pieces.append(child.text or "")
+            elif tag in (f"{{{W_NS}}}br", f"{{{W_NS}}}cr"):
+                pieces.append("\n")
+            elif tag == f"{{{W_NS}}}tab":
+                pieces.append("\t")
+        line = "".join(pieces).strip()
+        if line:
+            paragraphs.append(line)
+    return "\n".join(paragraphs)
+
+
+def _txt_text(raw: bytes) -> str:
+    # 中文 txt 常见三种落盘编码;utf-8 失败再退 GBK 家族,最后一步忽略坏字节保底出文本。
+    for encoding in ("utf-8-sig", "utf-8", "gb18030"):
+        try:
+            return raw.decode(encoding)
+        except UnicodeDecodeError:
+            continue
+    return raw.decode("utf-8", errors="ignore")
+
+
+def extract_script_text(upload) -> tuple[str, str]:
+    """(文件名, 正文)。只收 docx / txt,其余一律拒绝。"""
+    name = getattr(upload, "name", "") or "script"
+    lowered = name.lower()
+    if not lowered.endswith((".docx", ".txt")):
+        raise ScriptFileError("只支持 docx 和 txt 两种脚本文件")
+
+    raw = upload.read()
+    if len(raw) > MAX_UPLOAD_BYTES:
+        raise ScriptFileError("脚本文件不能超过 5MB")
+
+    text = _docx_text(raw) if lowered.endswith(".docx") else _txt_text(raw)
+    # 统一换行 + 压掉连续空行,避免原稿的排版空白撑爆后面的提示词
+    text = re.sub(r"\n{3,}", "\n\n", text.replace("\r\n", "\n").replace("\r", "\n")).strip()
+    if not text:
+        raise ScriptFileError("没能从这个文件里读到文字,请换一份")
+    return name, text
@@ -15,7 +15,13 @@ from rest_framework.viewsets import ModelViewSet

 from apps.ai.models import AITask, ModelConfig
 from apps.ai.providers import TtsNotConfigured
-from apps.ai.script_agent import stream_script_agent
+from apps.ai.script_agent import (
+    SEGMENT_DURATION_MAX,
+    SEGMENT_DURATION_MIN,
+    coerce_combo,
+    coerce_total_duration,
+    stream_script_agent,
+)
 from apps.ai.services import (
    DEFAULT_VOICEOVER_VOICE,
    VOICEOVER_VOICES,
@@ -39,6 +45,7 @@ from apps.assets.storage import TosStorage
 from apps.common.api import TeamScopedViewSetMixin
 from apps.common.celery_health import require_worker
 from apps.ai.generation_errors import classify_generation_error, public_error_for_task
+from apps.ai.video_digest import VideoDigestError, digest_project_video

 from .models import (
    BaseAssetGroup,
@@ -69,6 +76,7 @@ from .serializers import (
 )
 from .services.export import run_export_job_in_thread
 from .services.pipeline import STAGE_ORDER
+from .services.script_import import ScriptFileError, extract_script_text
 from .tasks import poll_video_segment_task

 logger = logging.getLogger(__name__)
@@ -417,14 +425,18 @@ class ProjectViewSet(TeamScopedViewSetMixin, ModelViewSet):
        project = serializer.save(team=self.get_team(), created_by=self.request.user)
        for stage in STAGE_ORDER:
            ProjectStage.objects.create(project=project, stage=stage)
+        # 先铺 4 段占位;真实段数与每段时长在采用脚本时由 _sync_video_segments_to_script 收口
        for index in range(4):
-            VideoSegment.objects.create(project=project, sort_order=index, target_duration_seconds=15)
+            VideoSegment.objects.create(
+                project=project, sort_order=index, target_duration_seconds=SEGMENT_DURATION_MAX
+            )

    @action(detail=True, methods=["post"], url_path="script-agent-stream", renderer_classes=[ServerSentEventRenderer])
    def script_agent_stream(self, request, pk=None):
        """对话式脚本 agent · 流式(SSE)。出稿 + 改稿一体,多模型可选。
        请求体:mode(auto|theme|revise)、prompt、model_config_id、selling_point_ids、
-        base_version_id(改稿)、aspect_ratio、total_duration。
+        base_version_id(改稿)、aspect_ratio、total_duration(5-60,5 秒步进)、
+        presentation_format(oral|drama|vlog)、video_structure(pain|contrast|review|scene)。
        响应:text/event-stream,逐帧吐 tool/delta/draft/saved/done/error。"""
        project = self.get_object()
        mode = str(request.data.get("mode") or "auto")
@@ -432,10 +444,12 @@ class ProjectViewSet(TeamScopedViewSetMixin, ModelViewSet):
        selling_point_ids = request.data.get("selling_point_ids") or []
        base_version_id = request.data.get("base_version_id") or None
        aspect_ratio = str(request.data.get("aspect_ratio") or "9:16")
-        try:
-            total_duration = int(request.data.get("total_duration") or 60)
-        except (TypeError, ValueError):
-            total_duration = 60
+        # 非法值一律由 agent 侧 coerce 兜底(夹区间/回落默认),这里不做 400,避免生成被参数噪声打断
+        total_duration = coerce_total_duration(request.data.get("total_duration"))
+        presentation_format, video_structure = coerce_combo(
+            request.data.get("presentation_format"),
+            request.data.get("video_structure"),
+        )
        target_index = request.data.get("target_index")
        try:
            target_index = int(target_index) if target_index is not None else None
@@ -466,7 +480,10 @@ class ProjectViewSet(TeamScopedViewSetMixin, ModelViewSet):
            base_version_id=base_version_id,
            aspect_ratio=aspect_ratio,
            total_duration=total_duration,
+            presentation_format=presentation_format,
+            video_structure=video_structure,
            target_index=target_index,
+            entry_source=str(request.data.get("source") or ""),
        )
        response = StreamingHttpResponse(stream, content_type="text/event-stream")
        response["Cache-Control"] = "no-cache"
@@ -790,11 +807,16 @@ class ProjectViewSet(TeamScopedViewSetMixin, ModelViewSet):
    # ── Stage 1 · 镜头脚本逐字段编辑 / 增删分镜 ──

    def _sync_video_segments_to_script(self, project: Project, script: ScriptVersion) -> None:
-        """采用版分镜数变化时,同步 VideoSegment 数量:不足则尾部补 NOT_STARTED,
-        多出且尾部是「从未生成过」的段则裁掉(已生成的段绝不动)。"""
+        """采用版分镜数变化时,同步 VideoSegment 的**数量和时长**:不足则尾部补 NOT_STARTED,
+        多出且尾部是「从未生成过」的段则裁掉(已生成的段绝不动)。
+
+        时长同步是二期补的:脚本镜可以不等长了,而出片、计价、时间线读的都是
+        VideoSegment.target_duration_seconds。不同步的话脚本写了 8 秒、出片仍按 15 秒跑。
+        已出片的段不改时长——改了会跟已渲染的成片对不上。"""
        if not script.is_adopted:
            return
-        target = script.segments.count()
+        script_segments = list(script.segments.order_by("sort_order"))
+        target = len(script_segments)
        segments = list(project.video_segments.order_by("sort_order"))
        while len(segments) > target:
            tail = segments[-1]
@@ -805,9 +827,34 @@ class ProjectViewSet(TeamScopedViewSetMixin, ModelViewSet):
                break
        next_order = (segments[-1].sort_order + 1) if segments else 0
        for _ in range(target - len(segments)):
-            segments.append(VideoSegment.objects.create(project=project, sort_order=next_order, target_duration_seconds=15))
+            index = len(segments)
+            seconds = (
+                script_segments[index].duration_seconds
+                if index < target
+                else SEGMENT_DURATION_MAX
+            )
+            segments.append(
+                VideoSegment.objects.create(
+                    project=project, sort_order=next_order, target_duration_seconds=seconds
+                )
+            )
            next_order += 1

+        # 已存在的段:只对「还没出过片」的回填脚本时长,已渲染的保持原样
+        stale: list[VideoSegment] = []
+        for index, video_segment in enumerate(segments):
+            if index >= target:
+                break
+            seconds = script_segments[index].duration_seconds
+            if not seconds or video_segment.target_duration_seconds == seconds:
+                continue
+            if video_segment.status == VideoSegment.Status.SUCCEEDED or video_segment.versions.exists():
+                continue
+            video_segment.target_duration_seconds = seconds
+            stale.append(video_segment)
+        if stale:
+            VideoSegment.objects.bulk_update(stale, ["target_duration_seconds"])
+
    def _sync_storyboard_shots_to_script(self, project: Project, script: ScriptVersion) -> None:
        """采用版分镜数变化时,同步 StoryboardShot 数量(与视频段同策略,按位置对齐):
        多出且尾部「从未出过图」的 shot 裁掉(已出图的不动);不主动建——出图时 ensure_storyboard_shots 按需补。"""
@@ -834,7 +881,10 @@ class ProjectViewSet(TeamScopedViewSetMixin, ModelViewSet):
                changed.append(field)
        if "duration_seconds" in request.data:
            try:
-                segment.duration_seconds = max(1, min(60, int(request.data["duration_seconds"])))
+                # 单镜必须落在 4–15 秒:出片模型的硬上限,越界下游直接拒片
+                segment.duration_seconds = max(
+                    SEGMENT_DURATION_MIN, min(SEGMENT_DURATION_MAX, int(request.data["duration_seconds"]))
+                )
                changed.append("duration_seconds")
            except (TypeError, ValueError):
                pass
@@ -877,7 +927,10 @@ class ProjectViewSet(TeamScopedViewSetMixin, ModelViewSet):
        created = ScriptSegment.objects.create(
            script_version=script,
            sort_order=insert_at,
-            duration_seconds=int(request.data.get("duration_seconds") or 15),
+            duration_seconds=max(
+                SEGMENT_DURATION_MIN,
+                min(SEGMENT_DURATION_MAX, int(request.data.get("duration_seconds") or SEGMENT_DURATION_MAX)),
+            ),
            narration=str(request.data.get("narration") or "").strip(),
            visual_prompt=str(request.data.get("visual_prompt") or "").strip(),
        )
@@ -1187,6 +1240,39 @@ class ProjectViewSet(TeamScopedViewSetMixin, ModelViewSet):
        segment.save(update_fields=["adopted_version", "status", "error_message", "updated_at"])
        return Response(ProjectSerializer(project).data, status=status.HTTP_201_CREATED)

+    @action(detail=True, methods=["post"], url_path="extract-script-file", parser_classes=[MultiPartParser, FormParser])
+    def extract_script_file_action(self, request, pk=None):
+        """上传脚本取正文:docx / txt → 纯文本。只读不落库,交前端填进对话框再走脚本 agent。"""
+        self.get_object()  # 走一遍团队权限校验
+        upload = request.data.get("file")
+        if upload is None:
+            return Response({"detail": "no file uploaded"}, status=status.HTTP_400_BAD_REQUEST)
+        try:
+            name, text = extract_script_text(upload)
+        except ScriptFileError as exc:
+            return Response({"detail": str(exc)}, status=status.HTTP_400_BAD_REQUEST)
+        return Response({"name": name, "chars": len(text), "text": text})
+
+    @action(detail=True, methods=["post"], url_path="extract-script-video", parser_classes=[MultiPartParser, FormParser])
+    def extract_script_video_action(self, request, pk=None):
+        """上传视频提炼:参考视频 → 中文分镜稿。
+
+        与 extract-script-file 同形状(只读不落脚本),交前端填进对话框让用户逐镜改,
+        改完再走脚本 agent 出结构化稿。差别是这条要真调模型,故按一次文本任务计费。
+        """
+        project = self.get_object()
+        upload = request.data.get("file")
+        if upload is None:
+            return Response({"detail": "no file uploaded"}, status=status.HTTP_400_BAD_REQUEST)
+        try:
+            result = digest_project_video(project=project, user=request.user, upload=upload)
+        except VideoDigestError as exc:
+            return Response({"detail": str(exc)}, status=status.HTTP_400_BAD_REQUEST)
+        except Exception as exc:  # noqa: BLE001 — 模型/网络失败:走统一安全文案,不回传原始异常
+            public_error = classify_generation_error(exc, operation="video_digest")
+            return Response(public_error, status=status.HTTP_502_BAD_GATEWAY)
+        return Response({"name": getattr(upload, "name", "") or "参考视频", **result})
+
    @action(detail=True, methods=["post"], url_path="upload-bgm", parser_classes=[MultiPartParser, FormParser])
    @transaction.atomic
    def upload_bgm_action(self, request, pk=None):