利用 AI 大模型和自动化工作流,根据主题或关键词一键生成高清短视频。Generate HD short videos from a topic or keyword with an automated AI workflow.
Python ai-video-generatorcontent-creationffmpeginstagram-reelsllmpythonshort-videosubtitlestext-to-speechtiktokvideo-automationvideo-workflowworkflow-automationyoutube-shorts
fix(subtitles): keep attached marks together during wrapping
Rudy Celekli676f91381996
diff --git a/app/services/video.py b/app/services/video.pyindex d131b3b6..485da337 100644--- a/app/services/video.py+++ b/app/services/video.py@@ -1327,13 +1327,22 @@ def wrap_text(text, max_width, font="Arial", fontsize=60):# 画布高度仍必须按现有行数计算,否则第二行及后续行会被裁掉。return text, (text.count("\n") + 1) * line_height+ def attached_text_units(token):+ units = []+ for char in token:+ if units and unicodedata.category(char).startswith("M"):+ units[-1] += char+ else:+ units.append(char)+ return units+def split_long_token(token):# 当一个 token 本身就超宽时(常见于中文无空格长句,或英文超长单词),# 退化为字符级拆分。关键点是:检测到 candidate 超宽时,先提交上一个# 仍然合法的 current,再把当前字符放入下一行,不能把超宽字符塞回上一行。lines = []current = ""- for char in token:+ for char in attached_text_units(token):candidate = f"{current}{char}"candidate_width, _ = get_text_size(candidate)if candidate_width <= max_width or not current:@@ -1379,11 +1388,15 @@ def wrap_text(text, max_width, font="Arial", fontsize=60):if len(lines[index - 1]) <= 1:continue- candidate = f"{lines[index - 1][-1]}{lines[index]}"+ previous_units = attached_text_units(lines[index - 1])+ if len(previous_units) <= 1:+ continue+ last_unit = previous_units[-1]+ candidate = f"{last_unit}{lines[index]}"candidate_width, _ = get_text_size(candidate)if candidate_width <= max_width:lines[index] = candidate- lines[index - 1] = lines[index - 1][:-1]+ lines[index - 1] = lines[index - 1][:-len(last_unit)]result = "\n".join(line.strip() for line in lines if line.strip()).strip()# 高度以最终结果为准。原文本中的显式换行可能保留在某个 token 内,diff --git a/test/services/test_subtitle_wrap_attached_marks.py b/test/services/test_subtitle_wrap_attached_marks.pynew file mode 100644index 00000000..190437fe--- /dev/null+++ b/test/services/test_subtitle_wrap_attached_marks.py@@ -0,0 +1,20 @@+import unicodedata+from pathlib import Path++import pytest+from PIL import ImageFont++from app.services.video import wrap_text++[email protected]("syllable", ["i\u0301", "i\u0308", "a\u0301"])+def test_subtitle_wrap_keeps_accent_with_its_base_character(syllable):+ font_path = Path(__file__).resolve().parents[2] / "resource/fonts/BeVietnamPro-Medium.ttf"+ text = syllable * 35+ wrapped, _ = wrap_text(text, 180, font=str(font_path), fontsize=60)+ lines = wrapped.splitlines()+ assert len(lines) > 1+ assert all(not unicodedata.category(line[0]).startswith("M") for line in lines)+ assert "".join(lines) == text+ font = ImageFont.truetype(str(font_path), 60)+ assert all(font.getbbox(line)[2] - font.getbbox(line)[0] <= 180 for line in lines)