feat: use image concurrency for outfit directories

src/app/widgets/ai_outfit_panel.py

@@ -112,6 +112,7 @@ class _OutfitWorker(QObject):
                 task, self._prompt, self._output_dir, self._model_config,
                 quality=self._quality, resolution=self._resolution,
                 request_interval=self._options.request_interval,
+                image_concurrency=self._options.concurrency,
                 image_log=self.log.emit,
             )
 
@@ -337,7 +338,7 @@ class AiOutfitPanel(QWidget):
         self._quality.addItems(_QUALITIES)
 
         pairs = [
-            ("并发数", self._concurrency), ("新请求间隔", self._interval),
+            ("图片并发数", self._concurrency), ("新请求间隔", self._interval),
             ("单任务冷却", self._cooldown), ("失败重试", self._retry_count),
             ("分辨率", self._resolution), ("JPG 质量", self._quality),
         ]

src/core/ai_outfit.py

@@ -1,7 +1,9 @@
 import logging
 import os
 import re
+import threading
 import time
+from concurrent.futures import ThreadPoolExecutor, as_completed
 from io import BytesIO
 from pathlib import Path
 
@@ -144,6 +146,7 @@ def generate_outfit_image(
     resolution="1K",
     api_client=None,
     request_interval=0.0,
+    image_concurrency=1,
     image_log=None,
 ):
     """Generate outfit image(s) for one Excel row and return OutfitResult.
@@ -160,7 +163,8 @@ def generate_outfit_image(
         return _generate_directory_outfit(
             task, prompt_template, output_dir, model_config,
             quality=quality, resolution=resolution, api_client=api_client,
-            request_interval=request_interval, image_log=image_log,
+            request_interval=request_interval, image_concurrency=image_concurrency,
+            image_log=image_log,
         )
 
     try:
@@ -198,6 +202,7 @@ def _generate_directory_outfit(
     resolution,
     api_client,
     request_interval,
+    image_concurrency,
     image_log,
 ):
     """Fan one directory row out into per-image generations (docs/11 §4.1/§9.1).
@@ -228,34 +233,43 @@ def _generate_directory_outfit(
     subdir_name = os.path.basename(os.path.normpath(directory))
     subdir_path = str(Path(output_dir) / safe_product_filename(subdir_name))
     prompt = render_prompt(prompt_template, task, resolution=resolution)
-    client = api_client or ImageApiClient(model_config)
     interval = float(request_interval or 0.0)
+    workers = max(1, int(image_concurrency or 1))
+    limiter = _RequestStartLimiter(interval)
+
+    def generate_one(index, image_path):
+        output_path = make_outfit_subdir_path(output_dir, subdir_name, image_path.stem)
+        if output_path.exists():
+            emit("第 {} 行 第 {}/{} 张已存在，跳过：{}".format(
+                task.row_index, index, total, image_path.name))
+            return True, str(output_path), ""
+        try:
+            limiter.wait()
+            client = api_client or ImageApiClient(model_config)
+            image_bytes = client.generate(prompt, str(image_path), resolution=resolution)
+            save_jpg_under_limit(image_bytes, output_path, quality=quality)
+            emit("第 {} 行 第 {}/{} 张完成：{}".format(
+                task.row_index, index, total, image_path.name))
+            return True, str(output_path), ""
+        except Exception as exc:                    # noqa: BLE001 - record and continue
+            emit("第 {} 行 第 {}/{} 张失败：{}（{}）".format(
+                task.row_index, index, total, image_path.name, exc))
+            return False, "", "{}：{}".format(image_path.name, exc)
 
     total = len(images)
     outputs = []
     failures = []
-    called = False
-    for index, image_path in enumerate(images, start=1):
-        output_path = make_outfit_subdir_path(output_dir, subdir_name, image_path.stem)
-        if output_path.exists():
-            outputs.append(str(output_path))
-            emit("第 {} 行 第 {}/{} 张已存在，跳过：{}".format(
-                task.row_index, index, total, image_path.name))
-            continue
-        try:
-            if called and interval > 0:
-                time.sleep(interval)
-            image_bytes = client.generate(prompt, str(image_path), resolution=resolution)
-            called = True
-            save_jpg_under_limit(image_bytes, output_path, quality=quality)
-            outputs.append(str(output_path))
-            emit("第 {} 行 第 {}/{} 张完成：{}".format(
-                task.row_index, index, total, image_path.name))
-        except Exception as exc:                    # noqa: BLE001 - record and continue
-            called = True
-            failures.append("{}：{}".format(image_path.name, exc))
-            emit("第 {} 行 第 {}/{} 张失败：{}（{}）".format(
-                task.row_index, index, total, image_path.name, exc))
+    with ThreadPoolExecutor(max_workers=workers) as executor:
+        futures = [
+            executor.submit(generate_one, index, image_path)
+            for index, image_path in enumerate(images, start=1)
+        ]
+        for future in as_completed(futures):
+            ok, output, error = future.result()
+            if ok:
+                outputs.append(output)
+            else:
+                failures.append(error)
 
     if failures:
         error = "{} 张中 {} 张失败：{}".format(total, len(failures), "；".join(failures))
@@ -278,6 +292,25 @@ def _coerce_quality(quality):
         return QUALITY_BALANCED
 
 
+class _RequestStartLimiter:
+    """Thread-safe minimum interval between AI request starts."""
+
+    def __init__(self, interval_seconds):
+        self.interval_seconds = max(0.0, float(interval_seconds or 0.0))
+        self._lock = threading.Lock()
+        self._next_time = 0.0
+
+    def wait(self):
+        if self.interval_seconds <= 0:
+            return
+        with self._lock:
+            now = time.monotonic()
+            if now < self._next_time:
+                time.sleep(self._next_time - now)
+                now = time.monotonic()
+            self._next_time = now + self.interval_seconds
+
+
 def _crop_square(img):
     width, height = img.size
     side = min(width, height)

src/core/outfit_batch.py

@@ -19,6 +19,8 @@ LogCallback = Callable[[str], None]
 class OutfitBatchOptions:
     """Batch generation controls for AI outfit tasks."""
 
+    # Kept as "concurrency" for app_config compatibility. Business meaning is
+    # now image concurrency inside one directory row; Excel row concurrency is 1.
     concurrency: int = 1
     request_interval: float = 0.0
     task_cooldown: float = 0.0
@@ -60,7 +62,7 @@ class RateLimiter:
 
 
 class OutfitBatchRunner:
-    """Run outfit tasks with concurrency, rate limiting, retries and stop."""
+    """Run Excel rows sequentially with retries, heartbeat and gentle stop."""
 
     def __init__(
         self,
@@ -80,11 +82,6 @@ class OutfitBatchRunner:
         self._sleep = sleep_func or time.sleep
         self._time = time_func or time.monotonic
         self._stop_event = threading.Event()
-        self._limiter = RateLimiter(
-            self.options.request_interval,
-            now_func=self._time,
-            sleep_func=self._sleep,
-        )
 
     def stop(self):
         """Request a gentle stop: no new tasks, in-flight tasks finish."""
@@ -94,11 +91,16 @@ class OutfitBatchRunner:
         total = len(self.tasks)
         results = []
         completed = 0
-        max_workers = max(1, int(self.options.concurrency or 1))
+        image_concurrency = max(1, int(self.options.concurrency or 1))
+        row_workers = 1
 
-        self._log("开始 AI 穿搭批量生成：{} 项，并发 {}".format(total, max_workers))
+        self._log(
+            "开始 AI 穿搭批量生成：{} 项，Excel 行并发 1，图片并发 {}".format(
+                total, image_concurrency
+            )
+        )
 
-        with ThreadPoolExecutor(max_workers=max_workers) as executor:
+        with ThreadPoolExecutor(max_workers=row_workers) as executor:
             task_iter = iter(self.tasks)
             futures = {}
             submitted_all = False
@@ -106,7 +108,7 @@ class OutfitBatchRunner:
             while futures or not submitted_all:
                 while (
                     not self._stop_event.is_set()
-                    and len(futures) < max_workers
+                    and len(futures) < row_workers
                     and not submitted_all
                 ):
                     try:
@@ -114,7 +116,6 @@ class OutfitBatchRunner:
                     except StopIteration:
                         submitted_all = True
                         break
-                    self._limiter.wait()
                     future = executor.submit(self._run_one_with_retry, task)
                     futures[future] = _RunningTask(
                         task=task,

tasks.md

@@ -1225,10 +1225,10 @@
 设计取舍：外层 `OutfitBatchRunner` 固定 Excel 行并发为 1；界面文案改为「图片并发数」。配置字段可先兼容复用现有 `outfit_concurrency`，但业务含义改为目录内图片并发；若后续改字段名，需要迁移旧配置。
 
 - [x] 文档更新：`docs/11-ai-outfit.md` 明确 Excel 行并发固定 1、图片并发数只作用于目录内部；`docs/ui-ai-outfit.html` 示例文案改为「图片并发数」
-- [ ] `ai_outfit_panel.py`：右侧生成设置 label 从「并发数」改为「图片并发数」，日志启动行同时显示「Excel 行并发 1 / 图片并发 N」
-- [ ] `_OutfitWorker` / `OutfitBatchRunner`：外层 Excel 行任务固定顺序处理，不再用界面并发值作为行 `max_workers`
-- [ ] `generate_outfit_image`：增加或接入 `image_concurrency` 参数；单文件行保持顺序单张处理
-- [ ] `core/ai_outfit.py` 目录分支：使用 `ThreadPoolExecutor(max_workers=图片并发数)` 并发生成目录内图片；已存在输出仍跳过，部分失败仍聚合到整行结果
-- [ ] 保留「新请求间隔」启动节流；图片并发数 >1 时日志/缩略图完成顺序允许与文件名排序不同
-- [ ] 补测试：Excel 行顺序处理、目录内图片并发、单文件行不并发、已存在跳过、部分失败聚合、UI label/default/config 兼容
-- [ ] 验证：相关单测通过；离屏启动 AI 穿搭页确认 label 和日志文案；如工作区清理后再跑全套测试
+- [x] `ai_outfit_panel.py`：右侧生成设置 label 从「并发数」改为「图片并发数」，日志启动行同时显示「Excel 行并发 1 / 图片并发 N」
+- [x] `_OutfitWorker` / `OutfitBatchRunner`：外层 Excel 行任务固定顺序处理，不再用界面并发值作为行 `max_workers`
+- [x] `generate_outfit_image`：增加或接入 `image_concurrency` 参数；单文件行保持顺序单张处理
+- [x] `core/ai_outfit.py` 目录分支：使用 `ThreadPoolExecutor(max_workers=图片并发数)` 并发生成目录内图片；已存在输出仍跳过，部分失败仍聚合到整行结果
+- [x] 保留「新请求间隔」启动节流；图片并发数 >1 时日志/缩略图完成顺序允许与文件名排序不同
+- [x] 补测试：Excel 行顺序处理、目录内图片并发、单文件行不并发、已存在跳过、部分失败聚合、UI label/default/config 兼容
+- [~] 验证：语法检查、`test_outfit_batch.py`、`test_ai_outfit.py`、`test_ai_outfit_panel.py` 通过；全套 `python -m unittest discover -s tests` 当前被工作区未提交的 `packaging/default_config/ai_models.json` 改动阻塞（模型顺序与出厂模板规范不一致），待清理该文件后重跑

tests/test_ai_outfit.py

@@ -3,6 +3,8 @@ import os
 import shutil
 import sys
 import tempfile
+import threading
+import time
 import unittest
 from io import BytesIO
 from pathlib import Path
@@ -262,6 +264,22 @@ class TestAiOutfitCore(unittest.TestCase):
         self.assertEqual(len(result.output_paths), 2)
         self.assertFalse((out / "a" / "img2.jpg").exists())
 
+    def test_generate_directory_uses_image_concurrency(self):
+        from core.ai_outfit import generate_outfit_image
+
+        d = self._make_dir_with_images("a", files=("img1.png", "img2.png", "img3.png", "img4.png"))
+        out = self.tmp / "out"
+        client = _ConcurrentRecordingClient(self._image_bytes())
+
+        result = generate_outfit_image(
+            self._dir_task(d), "x", out, model_config={}, api_client=client,
+            image_concurrency=2)
+
+        self.assertTrue(result.success, result.error)
+        self.assertEqual(client.calls, 4)
+        self.assertGreaterEqual(client.max_active, 2)
+        self.assertEqual(len(result.output_paths), 4)
+
 
 class _FakeClient:
     def __init__(self, image_bytes):
@@ -309,5 +327,26 @@ class _FailOnClient:
         return self._image_bytes
 
 
+class _ConcurrentRecordingClient:
+    """Thread-safe fake client that records concurrent generate() calls."""
+
+    def __init__(self, image_bytes):
+        self._image_bytes = image_bytes
+        self.calls = 0
+        self.active = 0
+        self.max_active = 0
+        self._lock = threading.Lock()
+
+    def generate(self, prompt, image_path, resolution="1K"):
+        with self._lock:
+            self.calls += 1
+            self.active += 1
+            self.max_active = max(self.max_active, self.active)
+        time.sleep(0.05)
+        with self._lock:
+            self.active -= 1
+        return self._image_bytes
+
+
 if __name__ == "__main__":
     unittest.main()

tests/test_ai_outfit_panel.py

@@ -64,6 +64,15 @@ class TestAiOutfitPanelDefaults(unittest.TestCase):
 
         self.assertEqual(Path(panel._output_edit.text()), saved)
 
+    def test_generation_setting_label_is_image_concurrency(self):
+        from PySide6.QtWidgets import QLabel
+
+        panel = self._panel()
+
+        labels = [label.text() for label in panel.findChildren(QLabel)]
+
+        self.assertIn("图片并发数", labels)
+
 
 if __name__ == "__main__":
     unittest.main()

tests/test_outfit_batch.py

@@ -91,15 +91,17 @@ class TestOutfitBatchRunner(unittest.TestCase):
         self.assertEqual(summary.success_count, 1)
         self.assertEqual(summary.results[0].attempts, 2)
 
-    def test_concurrency_runs_more_than_one_task_at_once(self):
+    def test_concurrency_option_does_not_parallelize_excel_rows(self):
         from core.outfit_batch import OutfitBatchOptions, OutfitBatchRunner
 
         active = [0]
         max_active = [0]
+        started = []
         lock = threading.Lock()
 
         def generate(task):
             with lock:
+                started.append(task.row_index)
                 active[0] += 1
                 max_active[0] = max(max_active[0], active[0])
             time.sleep(0.05)
@@ -116,7 +118,8 @@ class TestOutfitBatchRunner(unittest.TestCase):
         summary = runner.run()
 
         self.assertEqual(summary.success_count, 4)
-        self.assertGreaterEqual(max_active[0], 2)
+        self.assertEqual(max_active[0], 1)
+        self.assertEqual(started, [2, 3, 4, 5])
 
     def test_task_cooldown_is_applied(self):
         from core.outfit_batch import OutfitBatchOptions, OutfitBatchRunner