azmat1127/yingqing
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

优化故事版生成

Browse Source
This commit is contained in:
Azmat@qq.com
2026-08-21 18:22:50 +08:00
parent 158c66ac25
commit 409d1c2969
5 changed files with 181 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files
core/backend/apps/ai/services.py
+19 -2
View File
@@ -62,6 +62,23 @@ def get_default_model(capability: str) -> ModelConfig:
return qs.filter(is_default=True).order_by("created_at").first() or qs.order_by("created_at").first()
def get_storyboard_image_model() -> ModelConfig:
"""故事板出图钉 YunQi gpt-image-2 多图 edits(与手工测通的 curl 同一条链路)。
找不到再回落默认图像模型,避免测试/未 seed 环境直接挂。"""
pinned = (
ModelConfig.objects.select_related("provider")
.filter(
capability=ModelConfig.Capability.IMAGE,
status=ModelConfig.Status.ACTIVE,
provider__status="active",
provider__name="yunqi",
name="gpt-image-2",
)
.first()
)
return pinned or get_default_model(ModelConfig.Capability.IMAGE)
def resolve_image_model(key: str | None) -> "ModelConfig | None":
"""前端「生图模型选择」→ ModelConfig。用户显式选的可以是 disabled 模型(故不按 status 过滤)。
· "volcano" → 火山官方 Seedream(取最新一版)
@@ -2625,7 +2642,7 @@ def submit_storyboard(*, project, user, prompt: str = "", shot_ids: list | None
adopted_script = project.script_versions.filter(is_adopted=True).first()
if adopted_script is None:
raise ValueError("script must be adopted before generating storyboard")
if get_default_model(ModelConfig.Capability.IMAGE) is None:
if get_storyboard_image_model() is None:
raise ValueError("no active image model configured")
if prompt:
meta = dict(project.metadata or {})
@@ -3046,7 +3063,7 @@ def poll_storyboard(*, project, user) -> dict:
).values_list("request_payload__storyboard_shot", flat=True)
if v
}
model_config = get_default_model(ModelConfig.Capability.IMAGE)
model_config = get_storyboard_image_model()
extra_prompt = (project.metadata or {}).get("storyboard_prompt", "") or ""
spawnable = [s for s in active if str(s.id) not in inflight_shot_ids]
slots = max(0, STORYBOARD_MAX_PARALLEL - len(inflight_shot_ids))
core/backend/apps/ai/test_video_digest.py
+55 -1
View File
@@ -8,9 +8,11 @@ from __future__ import annotations
import subprocess
from django.core.files.uploadedfile import SimpleUploadedFile
from django.test import SimpleTestCase
from django.test import SimpleTestCase, TestCase
from apps.ai.models import ModelConfig, ModelProvider
from apps.ai.video_digest import (
DIGEST_VISION_MODEL_NAME,
MAX_DURATION_SECONDS,
MAX_FRAMES,
MAX_UPLOAD_BYTES,
@@ -22,6 +24,7 @@ from apps.ai.video_digest import (
frames_from_upload,
load_digest_skill,
plan_frame_times,
resolve_digest_model_config,
validate_digest_text,
)
@@ -121,6 +124,57 @@ class DigestValidationTests(SimpleTestCase):
self.assertEqual(validate_digest_text(f" {good} "), good.strip())
class DigestModelPinTests(TestCase):
"""视频提炼必须钉会看图的 Gemini,不能跟默认语言模型走。"""
def setUp(self):
ModelConfig.objects.filter(capability=ModelConfig.Capability.TEXT).update(
status=ModelConfig.Status.DISABLED
)
def _provider(self, name, display=""):
provider, _ = ModelProvider.objects.get_or_create(
name=name,
defaults={"display_name": display or name, "status": ModelProvider.Status.ACTIVE},
)
provider.display_name = display or name
provider.status = ModelProvider.Status.ACTIVE
provider.save(update_fields=["display_name", "status"])
return provider
def _model(self, provider, name, display_name, *, is_default=False):
model, _ = ModelConfig.objects.update_or_create(
provider=provider,
name=name,
capability=ModelConfig.Capability.TEXT,
defaults={
"display_name": display_name,
"status": ModelConfig.Status.ACTIVE,
"is_default": is_default,
},
)
return model
def test_prefers_official_relay_gemini_over_default_text_model(self):
doubao = self._provider("digest-pin-doubao", "火山")
relay = self._provider("digest-pin-relay", "Gemini 官转")
self._model(doubao, "doubao-seed-2-0-pro-260215", "豆包 2.0 Pro", is_default=True)
gemini = self._model(relay, DIGEST_VISION_MODEL_NAME, "Gemini 3.1 Pro 官转")
picked = resolve_digest_model_config()
self.assertEqual(picked.id, gemini.id)
def test_falls_back_to_gemini_name_when_label_has_no_relay_tag(self):
yunqi = self._provider("digest-pin-yunqi", "YunQi · Gemini")
gemini = self._model(yunqi, DIGEST_VISION_MODEL_NAME, "Gemini 3.1 Pro")
picked = resolve_digest_model_config()
self.assertEqual(picked.id, gemini.id)
def test_does_not_use_default_language_model_if_gemini_missing(self):
doubao = self._provider("digest-pin-text-only", "火山")
self._model(doubao, "doubao-seed-2-0-pro-260215", "豆包 2.0 Pro", is_default=True)
self.assertIsNone(resolve_digest_model_config())
# --------------------------------------------------------------------------- #
# helpers
# --------------------------------------------------------------------------- #
core/backend/apps/ai/video_digest.py
+60 -6
View File
@@ -3,8 +3,9 @@
链路:ffmpeg 抽帧(均匀采样) → 帧内联进多模态 messages → 走现有文本模型路由 → 纯文本分镜稿。
两个「本来以为要新建、其实已经有」的前提:
1. **读图能力**:默认脚本模型(YunQi gemini-3.1-pro)本身是多模态的,OpenAI 兼容的
``content: [{type:"text"},{type:"image_url"}]`` 直接透传即可,不需要新 provider、新模型、新 key。
1. **读图能力**:拆视频是多模态(帧图 + 文本)。默认语言模型现在可能是纯文本豆包,
**不能再用 get_default_model(TEXT)**。固定钉 Gemini 3.1 Pro 官转
(``gemini-3.1-pro-preview``,展示名带「官转」优先)。
2. **ffmpeg**:第 5 阶段导出早就依赖它,已装进后端镜像(见 Dockerfile)。
**没有音轨**:帧里读不到口播,只能读画面上的字幕。这是刻意取舍——接语音转写要另开火山 ASR 服务、
@@ -198,6 +199,59 @@ def validate_digest_text(text: str) -> str:
return cleaned
# 拆视频必须会看图。默认文本模型现在常是纯文本豆包,传帧图会直接失败。
# 钉 Gemini 3.1 Pro;展示名带「官转」的优先(中转站官方 Gemini 通道)。
DIGEST_VISION_MODEL_NAME = "gemini-3.1-pro-preview"
def resolve_digest_model_config():
"""视频提炼用的多模态文本模型:Gemini 3.1 Pro 官转。找不到不回落默认语言模型。"""
from apps.ai.models import ModelConfig
qs = (
ModelConfig.objects.select_related("provider")
.filter(
capability=ModelConfig.Capability.TEXT,
status=ModelConfig.Status.ACTIVE,
provider__status="active",
)
)
def _blob(model) -> str:
return " ".join(
filter(
None,
[
model.name,
model.display_name,
getattr(model.provider, "name", ""),
getattr(model.provider, "display_name", ""),
],
)
)
ranked = []
for model in qs:
blob = _blob(model)
name_hit = "gemini-3.1" in model.name.lower() or "gemini-3.1" in (model.display_name or "").lower()
label_hit = "gemini 3.1" in (model.display_name or "").lower()
if not (name_hit or label_hit or model.name == DIGEST_VISION_MODEL_NAME):
continue
# 官转 > 精确模型名 > 其它 Gemini 3.1
score = 0
if "官转" in blob:
score += 100
if model.name == DIGEST_VISION_MODEL_NAME:
score += 20
if getattr(model.provider, "name", "") == "yunqi_gemini":
score += 5
ranked.append((score, model.created_at, model))
if not ranked:
return None
ranked.sort(key=lambda item: (-item[0], item[1]))
return ranked[0][2]
# --------------------------------------------------------------------------- #
# 入口:一次真实的计费调用
# --------------------------------------------------------------------------- #
@@ -206,15 +260,15 @@ def digest_project_video(*, project, user, upload) -> dict:
from django.db import transaction
from django.utils import timezone
from apps.ai.models import AITask, ModelConfig
from apps.ai.services import create_ai_task, execute_routed_text_request, get_default_model
from apps.ai.models import AITask
from apps.ai.services import create_ai_task, execute_routed_text_request
from apps.billing.services.ledger import charge_reserved_credit
frames, duration = frames_from_upload(upload)
model_config = get_default_model(ModelConfig.Capability.TEXT)
model_config = resolve_digest_model_config()
if model_config is None:
raise VideoDigestError("暂时没有可用的模型,请联系管理员")
raise VideoDigestError("视频提炼需要 Gemini 3.1 Pro(会看图),当前没有启用,请联系管理员")
product = getattr(project, "product", None)
messages = build_digest_messages(
core/frontend/src/pipeline-page.css
+40 -2
View File
@@ -352,7 +352,7 @@
.del { text-decoration: line-through; color: var(--black-alpha-48); }
.ins { background: var(--forest-bg); color: var(--accent-forest); padding: 0 3px; }
.chat-input { padding: 14px 18px 18px; border-top: 1px solid var(--border-faint); }
.chat-input-card { background: var(--background-base); border: 1px solid var(--border-faint); border-radius: var(--r-md); padding: 12px 14px 10px; transition: border-color var(--t-base), box-shadow var(--t-base); }
.chat-input-card { background: var(--background-base); border: 1px solid var(--border-faint); border-radius: var(--r-md); padding: 12px 14px 10px; min-width: 0; overflow: hidden; transition: border-color var(--t-base), box-shadow var(--t-base); }
.chat-input-card:focus-within { border-color: rgba(0, 47, 167, 0.38); box-shadow: 0 0 0 3px rgba(0, 47, 167, 0.07); }
.chat-input-area { width: 100%; border: none; outline: none; background: transparent; font-family: var(--font-sans); font-size: 13px; color: var(--accent-black); line-height: 1.55; resize: none; padding: 0; min-height: 42px; }
.chat-input-area::placeholder { color: var(--black-alpha-40); }
@@ -367,7 +367,45 @@
.chat-send-btn:hover { background: var(--klein-hover); border-color: var(--klein-hover); }
.chat-send-btn:active { transform: scale(.95); }
.chat-send-btn:disabled { background: var(--black-alpha-12); border-color: var(--black-alpha-12); color: var(--black-alpha-40); cursor: not-allowed; transform: none; }
.chat-attach-row { display: flex; flex-wrap: wrap; gap: 6px; margin-bottom: 8px; }
.chat-attach-row { display: flex; flex-wrap: wrap; gap: 6px; margin-bottom: 8px; min-width: 0; max-width: 100%; }
.chat-file {
display: flex;
align-items: center;
gap: 8px;
max-width: 100%;
min-width: 0;
padding: 8px 10px;
background: var(--surface);
border: 1px solid var(--border-faint);
border-radius: var(--r-md);
}
.chat-file-ico { width: 16px; height: 16px; flex-shrink: 0; color: var(--black-alpha-48); }
.chat-file-meta { min-width: 0; flex: 1; display: flex; flex-direction: column; gap: 1px; }
.chat-file-name {
overflow: hidden;
text-overflow: ellipsis;
white-space: nowrap;
font-size: 13px;
color: var(--accent-black);
line-height: 1.3;
}
.chat-file-sub { font-size: 12px; color: var(--black-alpha-48); line-height: 1.3; }
.chat-file-x {
flex-shrink: 0;
width: 18px;
height: 18px;
display: grid;
place-items: center;
background: transparent;
border: 0;
border-radius: var(--r-sm);
color: var(--black-alpha-48);
cursor: pointer;
font-size: 14px;
line-height: 1;
padding: 0;
}
.chat-file-x:hover { background: var(--black-alpha-4); color: var(--accent-black); }
.shot-list { display: flex; flex-direction: column; }
.shots-body { padding: 16px 18px; flex: 1; overflow-y: auto; max-height: 540px; display: flex; flex-direction: column; gap: 12px; position: relative; align-items: stretch; }
core/frontend/src/routes/pipeline.tsx
+7 -4
View File
@@ -3157,10 +3157,13 @@ export function PipelinePage(props: {
</div>
<div className="chat-attach-row" id="chat-attach-row" hidden={chatAttachments.length === 0}>
{chatAttachments.map((att, index) => (
<span className="chip" key={`${att.name}-${index}`} style={{ marginRight: 6 }}>
<svg width="11" height="11" viewBox="0 0 24 24" fill="none" stroke="currentColor" strokeWidth="1.6" strokeLinecap="round" strokeLinejoin="round" style={{ marginRight: 4 }}><path d="M14 2H6a2 2 0 0 0-2 2v16a2 2 0 0 0 2 2h12a2 2 0 0 0 2-2V8z" /><path d="M14 2v6h6" /></svg>
{att.name} · {att.chars} 字
<button type="button" aria-label="移除附件" style={{ marginLeft: 4, background: "none", border: 0, cursor: "pointer", color: "inherit" }} onClick={() => setChatAttachments((list) => list.filter((_, i) => i !== index))}>×</button>
<span className="chat-file" key={`${att.name}-${index}`}>
<svg className="chat-file-ico" viewBox="0 0 24 24" fill="none" stroke="currentColor" strokeWidth="1.6" strokeLinecap="round" strokeLinejoin="round" aria-hidden="true"><path d="M14 2H6a2 2 0 0 0-2 2v16a2 2 0 0 0 2 2h12a2 2 0 0 0 2-2V8z" /><path d="M14 2v6h6" /></svg>
<span className="chat-file-meta">
<span className="chat-file-name" title={att.name}>{att.name}</span>
<span className="chat-file-sub">{att.chars} 字</span>
</span>
<button className="chat-file-x" type="button" aria-label="移除附件" onClick={() => setChatAttachments((list) => list.filter((_, i) => i !== index))}>×</button>
</span>
))}
</div>