nex_animation/backend/tests/test_shot_preview.py

212 lines
11 KiB
Python
Raw Permalink Blame History

This file contains ambiguous Unicode characters!

This file contains ambiguous Unicode characters that may be confused with others in your current locale. If your use case is intentional and legitimate, you can safely ignore this warning. Use the Escape button to highlight these characters.

"""镜头「审片预览」平台能力 + 模型配置两处静默坑位的回归测试。
对应验收中发现的两个问题:
1. 音画同出的模型(MiniMax H3)自带一条即兴音轨,评审在分镜/审核页听到的
是模型原声而不是平台配音,被误判成「配音错了」;
现在默认口径已反过来:有台词的镜头优先用模型原声,TTS 只在显式 dub 或
模型不会出声时兜底,所以本文件的配音用例都先 `_dub_mode()` 显式切换;
2. PATCH /api/providers/{id} 传部分 config 会把整份配置覆盖掉;
config.megapixels 调小后画面静默降档,预检只比对声明档位所以不报警。
"""
from __future__ import annotations
import subprocess
from pathlib import Path
import pytest
from app.services import ffmpeg as ff, storage
from tests.test_delivery import _make_wav
def _make_video_with_audio(path: Path, *, seconds: float = 2.0, freq: float = 180.0,
size: str = "320x180", audio: bool = True) -> Path:
"""造一条真能解码、且带模型原声的镜头产物(模拟音画同出模型)。"""
if not ff.available():
pytest.skip("本机缺少 ffmpeg,无法验证审片预览")
path.parent.mkdir(parents=True, exist_ok=True)
cmd = [ff.FFMPEG, "-v", "error", "-y",
"-f", "lavfi", "-i", f"testsrc=size={size}:rate=24:duration={seconds}",
"-f", "lavfi", "-i", f"sine=frequency={freq}:duration={seconds}",
"-t", f"{seconds}", "-c:v", "libx264", "-preset", "veryfast", "-pix_fmt", "yuv420p"]
cmd += ["-c:a", "aac", "-b:a", "96k", "-shortest"] if audio else ["-an"]
cmd += [str(path)]
subprocess.run(cmd, check=True, capture_output=True)
return path
@pytest.fixture
def shot_with_output(api, studio):
"""一个「有模型产物(自带原声)+ 有平台配音」的镜头。"""
pid = studio["project_id"]
sid = api.post(f"/api/projects/{pid}/shots",
json={"title": "雾大", "description": "浓雾渡口,老船夫按住船舷", "mode": "i2v",
"duration": 2, "dialogue": "雾大,站稳再下船。"})["result"]["id"]
shot = api.get(f"/api/projects/{pid}/shots/{sid}")["shot"]
out = storage.project_dir(pid, "generated") / f"shot-native-{sid}.mp4"
_make_video_with_audio(out)
voice = storage.project_dir(pid, "voiceover") / f"shot-{sid}.wav"
voice.write_bytes(_make_wav(1.2, 520.0))
return {"pid": pid, "sid": sid, "out": out, "voice": voice, "code": shot["code"]}
def _attach(api, db, ctx):
from app.models import Shot
shot = db.get(Shot, ctx["sid"])
shot.output_path = storage.to_relative(ctx["out"])
shot.voiceover_path = storage.to_relative(ctx["voice"])
shot.voiceover_duration_ms = 1200
db.commit()
def _dub_mode(api, ctx):
"""把镜头切到「平台配音」口径。
出厂默认已经改成模型原声优先(voice_source=native,原生音画同步的产物
不再被 TTS 盖掉),而这几条用例验证的正是「换嗓子」那条路:必须显式 dub,
否则测的是新默认而不是配音替换。
"""
api.post(f"/api/projects/{ctx['pid']}/timeline",
json={"only_approved": False, "settings": {"voice_source": "dub"}})
# ---------------------------------------------------------------- 审片预览
def test_preview_replaces_model_audio_with_platform_voiceover(api, db, shot_with_output):
ctx = shot_with_output
_attach(api, db, ctx)
_dub_mode(api, ctx)
before = api.get(f"/api/projects/{ctx['pid']}/shots/{ctx['sid']}")["shot"]
assert before["native_audio"] is True and before["voiceover_ready"] is True
assert before["preview_ready"] is False and before["preview_url"] == ""
built = api.post(f"/api/projects/{ctx['pid']}/shots/{ctx['sid']}/preview",
json={"force": False, "actor": "acceptance"})
assert built["ok"] is True and built["preview"]["reused"] is False
assert built["preview"]["mode"] == "voiceover"
preview_path = storage.to_absolute(built["preview"]["path"])
assert preview_path.exists()
probed = ff.probe(preview_path)
video = ff.probe(shot_with_output["out"])
assert probed.has_audio is True
# 画面直接 copy:不重编码、不降档,尺寸与模型产物完全一致
assert (probed.width, probed.height) == (video.width, video.height)
assert probed.codec == video.codec
# 音轨与画面等长(尾部补静音),入点留白不会把画面截短
assert abs(probed.duration - video.duration) < 0.25
after = api.get(f"/api/projects/{ctx['pid']}/shots/{ctx['sid']}")["shot"]
assert after["preview_ready"] is True and after["preview_url"].startswith("/media/")
assert after["preview_url"].endswith(".mp4")
# 分镜看板同样带出预览,列表页不用逐条再点详情
board = api.get(f"/api/projects/{ctx['pid']}/shots")["shots"]
row = next(r for r in board if r["id"] == ctx["sid"])
assert row["preview_ready"] is True and row["preview_url"] == after["preview_url"]
# 预览产物登记为镜头资产,走统一媒体抽象
asset = next(a for a in after["assets"] if a["kind"] == "preview_video")
assert asset["path"] == built["preview"]["path"]
def test_preview_is_idempotent_and_rebuilt_when_inputs_change(api, db, shot_with_output):
ctx = shot_with_output
_attach(api, db, ctx)
_dub_mode(api, ctx)
url = f"/api/projects/{ctx['pid']}/shots/{ctx['sid']}/preview"
first = api.post(url, json={})["preview"]
again = api.post(url, json={})["preview"]
assert again["reused"] is True and again["path"] == first["path"]
# 换配音 → 指纹变化,必须重新封装(否则改台词后听到的还是旧声音)
ctx["voice"].write_bytes(_make_wav(1.6, 660.0))
rebuilt = api.post(url, json={})["preview"]
assert rebuilt["reused"] is False and rebuilt["path"] != first["path"]
forced = api.post(url, json={"force": True})["preview"]
assert forced["reused"] is False
batch = api.post(f"/api/projects/{ctx['pid']}/previews", json={})
assert batch["project_id"] == ctx["pid"] and batch["result"]["reused"] >= 1
def test_preview_falls_back_when_shot_has_no_voiceover(api, db, shot_with_output):
ctx = shot_with_output
_attach(api, db, ctx)
_dub_mode(api, ctx)
from app.models import Shot
shot = db.get(Shot, ctx["sid"])
shot.voiceover_path = ""
db.commit()
result = api.post(f"/api/projects/{ctx['pid']}/shots/{ctx['sid']}/preview", json={})
assert result["ok"] is True and result["preview"]["mode"] == "native"
detail = api.get(f"/api/projects/{ctx['pid']}/shots/{ctx['sid']}")["shot"]
assert detail["preview_ready"] is False and detail["voiceover_ready"] is False
def test_preview_without_output_is_a_clear_400(api, studio):
pid = studio["project_id"]
sid = api.post(f"/api/projects/{pid}/shots",
json={"title": "还没生成", "description": "等待生成的镜头", "mode": "i2v",
"duration": 2})["result"]["id"]
body = api.call("POST", f"/api/projects/{pid}/shots/{sid}/preview", json={}, expect=400)
assert "生成视频" in body["detail"]
# ---------------------------------------------------------------- 配置覆盖坑
def test_provider_config_patch_merges_by_default(api):
created = api.post("/api/providers", json={"provider": "localtest", "model": "cfg-merge",
"media_type": "video", "adapter": "mock",
"config": {"base_url": "http://127.0.0.1:9199",
"workflow": "minimax_h3",
"megapixels": 0.4}})["id"]
api.patch(f"/api/providers/{created}", json={"config": {"megapixels": 0.98}})
config = api.get("/api/providers")["providers"]
spec = next(p for p in config if p["id"] == created)
assert spec["config"]["megapixels"] == pytest.approx(0.98)
# 关键断言:只传一个键不得清空 base_url / workflow(旧实现会,且仍返回 200)
assert spec["config"]["base_url"] == "http://127.0.0.1:9199"
assert spec["config"]["workflow"] == "minimax_h3"
api.patch(f"/api/providers/{created}", json={"config": {"only_key": 1}, "replace_config": True})
spec = next(p for p in api.get("/api/providers")["providers"] if p["id"] == created)
assert spec["config"] == {"only_key": 1}
def test_probe_reports_effective_canvas(api):
created = api.post("/api/providers", json={"provider": "localtest", "model": "cfg-canvas",
"media_type": "video", "adapter": "comfyui",
"config": {"base_url": "http://127.0.0.1:1",
"workflow": "minimax_h3",
"megapixels": 0.4}})["id"]
probe = api.get(f"/api/providers/{created}/probe")
assert probe["canvas"]["16:9"] == [832, 480] # 声明 720p,实际只出 480p
api.patch(f"/api/providers/{created}", json={"config": {"megapixels": 0.98}})
assert api.get(f"/api/providers/{created}/probe")["canvas"]["16:9"] == [1312, 736]
def test_preflight_warns_when_effective_canvas_is_silently_downgraded(api, studio):
pid = studio["project_id"]
spec = api.post("/api/providers", json={"provider": "localtest", "model": "cfg-low",
"media_type": "video", "adapter": "comfyui",
"resolutions": ["480p", "720p"],
"config": {"base_url": "http://127.0.0.1:1",
"workflow": "minimax_h3",
"megapixels": 0.4}})["id"]
api.patch(f"/api/projects/{pid}", json={"resolution": "720p", "default_provider": "localtest",
"default_model": "cfg-low"})
api.post(f"/api/projects/{pid}/shots",
json={"title": "低像素", "description": "浓雾里的渡口灯火", "mode": "t2v", "duration": 6})
codes = {i["code"]: i for c in api.get(f"/api/projects/{pid}/preflight")["preflight"]["checks"]
for i in c["issues"]}
assert "resolution_effective_lower" in codes, codes
assert "832x480" in codes["resolution_effective_lower"]["message"]
# 把像素数调回真实 720p 档 → 警告消失(档位声明没变,只有配置变)
api.patch(f"/api/providers/{spec}", json={"config": {"megapixels": 0.98}})
codes = {i["code"] for c in api.get(f"/api/projects/{pid}/preflight")["preflight"]["checks"]
for i in c["issues"]}
assert "resolution_effective_lower" not in codes