{"data":{"base":{"version":"2026.08","eyebrow":"AI 能做什么","title":"推荐魔方","faces":[{"id":"capability","label":"能力","note":"先看能做什么；切换其他面时，同一输入输出位置保持不变。","rotation":{"x":0,"y":0}},{"id":"model","label":"模型","note":"只比较底层模型的任务长板，不把产品体验混在一起。","rotation":{"x":0,"y":-90}},{"id":"tool","label":"工具","note":"看普通用户实际打开什么，才能更省事地交付结果。","rotation":{"x":0,"y":-180}},{"id":"free","label":"免费","note":"看无需付费或可免费试用的入口；发布前必须复核额度与日期。","rotation":{"x":0,"y":90}},{"id":"domestic","label":"国内","note":"看在中国大陆更方便直接使用的方案，不等同于国产模型。","rotation":{"x":-90,"y":0}},{"id":"local","label":"本地","note":"按隐私、离线、低延迟和批处理价值推荐，不以编程跑分为中心。","rotation":{"x":90,"y":0}}],"media":[{"id":"text","label":"文","longLabel":"文字"},{"id":"image","label":"图","longLabel":"图片"},{"id":"video","label":"视","longLabel":"视频"},{"id":"audio","label":"音","longLabel":"音频"}],"transforms":[{"id":"text:text","input":"text","output":"text","children":[{"id":"text:text:writing","label":"写作/改写"},{"id":"text:text:translation","label":"翻译"},{"id":"text:text:research","label":"查找核对"},{"id":"text:text:coding","label":"编码工作"},{"id":"text:text:math","label":"数学"}]},{"id":"text:image","input":"text","output":"image","children":[{"id":"text:image:drawing","label":"绘图"},{"id":"text:image:design","label":"视觉设计"},{"id":"text:image:chart","label":"图表"}]},{"id":"text:video","input":"text","output":"video","children":[{"id":"text:video:generation","label":"文生视频"},{"id":"text:video:storyboard","label":"分镜预演"},{"id":"text:video:sync-audio","label":"同步声画生成"}]},{"id":"text:audio","input":"text","output":"audio","children":[{"id":"text:audio:voice","label":"配音"},{"id":"text:audio:music","label":"作曲"},{"id":"text:audio:effects","label":"音效"}]},{"id":"image:text","input":"image","output":"text","children":[{"id":"image:text:ocr","label":"读图识字"},{"id":"image:text:document","label":"表格 / Word/PDF 还原"},{"id":"image:text:qa","label":"看图问答"},{"id":"image:text:geolocation","label":"图片定位"}]},{"id":"image:image","input":"image","output":"image","children":[{"id":"image:image:retouch","label":"修图"},{"id":"image:image:redraw","label":"局部重绘"},{"id":"image:image:style","label":"风格转换"},{"id":"image:image:cutout","label":"去背景"}]},{"id":"image:video","input":"image","output":"video","children":[{"id":"image:video:generation","label":"图生视频"},{"id":"image:video:character","label":"角色动画"},{"id":"image:video:sync-audio","label":"同步声画生成"}]},{"id":"image:audio","input":"image","output":"audio","children":[{"id":"image:audio:narration","label":"看图讲述"},{"id":"image:audio:accessibility","label":"无障碍描述"},{"id":"image:audio:music","label":"图片配乐"}]},{"id":"video:text","input":"video","output":"text","children":[{"id":"video:text:subtitle","label":"字幕转写"},{"id":"video:text:summary","label":"内容总结"},{"id":"video:text:qa","label":"视频问答"},{"id":"video:text:course","label":"课程笔记"}]},{"id":"video:image","input":"video","output":"image","children":[{"id":"video:image:frame","label":"定点 / 等间隔抽帧"},{"id":"video:image:semantic-frame","label":"语义抽帧"},{"id":"video:image:cover-frame","label":"封面选帧"},{"id":"video:image:thumbnail","label":"缩略图重设计"},{"id":"video:image:storyboard","label":"分镜提取"}]},{"id":"video:video","input":"video","output":"video","children":[{"id":"video:video:supplement","label":"素材补充"},{"id":"video:video:remix","label":"二创"},{"id":"video:video:edit","label":"剪辑"},{"id":"video:video:translation","label":"翻译"},{"id":"video:video:restore","label":"修复 / 超分"}]},{"id":"video:audio","input":"video","output":"audio","children":[{"id":"video:audio:extract","label":"音轨提取"},{"id":"video:audio:dub","label":"重新配音"},{"id":"video:audio:effects","label":"视频配音效 / Foley"}]},{"id":"audio:text","input":"audio","output":"text","children":[{"id":"audio:text:transcribe","label":"语音转写"},{"id":"audio:text:minutes","label":"会议纪要"},{"id":"audio:text:lyrics","label":"歌词识别"},{"id":"audio:text:diarization","label":"说话人分离 / 标注"}]},{"id":"audio:image","input":"audio","output":"image","children":[{"id":"audio:image:cover","label":"专辑封面"},{"id":"audio:image:waveform","label":"波形"},{"id":"audio:image:spectrum","label":"频谱图"}]},{"id":"audio:video","input":"audio","output":"video","children":[{"id":"audio:video:mv","label":"MV"},{"id":"audio:video:lip-sync","label":"口型"},{"id":"audio:video:avatar","label":"数字人"}]},{"id":"audio:audio","input":"audio","output":"audio","children":[{"id":"audio:audio:denoise","label":"降噪"},{"id":"audio:audio:separate","label":"分轨"},{"id":"audio:audio:voice","label":"变声"},{"id":"audio:audio:continue","label":"续作"},{"id":"audio:audio:translation","label":"语音翻译"}]}]},"face":{"faceId":"capability","transforms":[{"id":"text:text","facet":{"shortLabel":"写作 / 检索","items":["写作改写","翻译","查找核对","编码工作","数学"],"detail":"AI生成始终受水印约束，人类看得多了，就能闻到AI味了\n\n代码编写能力，日后无人能敌","status":"ready","updatedAt":"2026-08-23"},"children":[{"id":"text:text:writing","facet":{"shortLabel":"L3 熟练级","items":["WritingBench / EditEval：写作与指令改写测评"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":3,"references":[{"title":"WritingBench","url":"https://arxiv.org/abs/2503.05244"},{"title":"EditEval","url":"https://aclanthology.org/2024.conll-1.7/"}]}},{"id":"text:text:translation","facet":{"shortLabel":"L4 职业级","items":["WMT25 General MT：多语言人工评价与系统排名"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":4,"references":[{"title":"WMT25 General MT","url":"https://github.com/wmt-conference/wmt25-general-mt"}]}},{"id":"text:text:research","facet":{"shortLabel":"L3 熟练级","items":["DeepResearch Bench II：最强系统的专家 rubric 满足率仍低于 50%"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":3,"references":[{"title":"DeepResearch Bench II","url":"https://arxiv.org/abs/2601.08536"}]}},{"id":"text:text:coding","facet":{"shortLabel":"L5 专家协作级","items":["SWE-rebench：真实仓库任务的 Result@1 与多次运行结果"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":5,"references":[{"title":"SWE-rebench","url":"https://swe-rebench.com/about"}]}},{"id":"text:text:math","facet":{"shortLabel":"L5 专家协作级","items":["AI4Math周报：Yau-Tian-Donaldson猜想被证伪后"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":5,"references":[{"title":"Disproof of the Yau--Tian--Donaldson conjecture","url":"https://arxiv.org/abs/2608.19301"},{"title":"AI4Math周报：Yau-Tian-Donaldson猜想被证伪后","url":"https://www.bilibili.com/video/BV1mX886VEtT/"},{"title":"【每日核爆】Yau-Tian-Donaldson猜想被推翻？！","url":"https://www.bilibili.com/video/BV1xF8M6gEFH/"},{"title":"一线科研人员使用AI一个月后的体验：AI到底有多强？数学家会被取代吗？谈谈我的看法","url":"https://www.bilibili.com/video/BV1C8836jE4N/"}]}}]},{"id":"text:image","facet":{"shortLabel":"绘图 / 设计","items":["绘图","视觉设计","图表"],"detail":"顶尖图像模型已经能理解自然中文，不必为了显得专业而强行套用 JSON 提示词。","status":"ready","updatedAt":"2026-08-23"},"children":[{"id":"text:image:drawing","facet":{"shortLabel":"L3 熟练级","items":["GenEval：对象、数量、颜色与位置组合"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":3,"references":[{"title":"GenEval","url":"https://proceedings.neurips.cc/paper_files/paper/2023/file/a3bf71c7c63f0c3bcb7ff67c67b1e7b1-Paper-Datasets_and_Benchmarks.pdf"}]}},{"id":"text:image:design","facet":{"shortLabel":"L2 助手级","items":["GraphicDesignBench：布局、字体、矢量与动画等职业设计核心能力仍明显不足"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":2,"references":[{"title":"GraphicDesignBench","url":"https://arxiv.org/abs/2604.04192"}]}},{"id":"text:image:chart","facet":{"shortLabel":"L2 助手级","items":["Text2Vis：1,985 个样本、20 余种图表，评估代码执行与图表质量"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":2,"references":[{"title":"Text2Vis","url":"https://aclanthology.org/2025.emnlp-main.1622/"}]}}]},{"id":"text:video","facet":{"shortLabel":"文生视频","items":["文生视频","分镜预演","同步声画生成"],"detail":"目前更接近二创与抽卡：脚本、分镜和后期剪辑仍然决定成片质量。","status":"ready","updatedAt":"2026-08-23"},"children":[{"id":"text:video:generation","facet":{"shortLabel":"L2 助手级","items":["VBench：16 维视频生成质量与人类对齐评测"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":2,"references":[{"title":"VBench","url":"https://github.com/Vchitect/VBench"}]}},{"id":"text:video:storyboard","facet":{"shortLabel":"L2 助手级","items":["Rich Storyboard Benchmark：提示对齐、角色一致性、场景多样性与用户研究"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":2,"references":[{"title":"Rich Storyboard Benchmark","url":"https://arxiv.org/abs/2508.09983"}]}},{"id":"text:video:sync-audio","facet":{"shortLabel":"L2 助手级","items":["VABench / AVGen-Bench：文本生成同步声画"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":2,"references":[{"title":"VABench","url":"https://openaccess.thecvf.com/content/CVPR2026/html/Hua_VABench_A_Comprehensive_Benchmark_for_Audio-Video_Generation_CVPR_2026_paper.html"},{"title":"AVGen-Bench","url":"https://github.com/microsoft/AVGen-Bench"}]}}]},{"id":"text:audio","facet":{"shortLabel":"配音 / 作曲","items":["配音","作曲","音效"],"detail":"","status":"ready","updatedAt":"2026-08-23"},"children":[{"id":"text:audio:voice","facet":{"shortLabel":"L4 职业级","items":["Seed-TTS Eval：中英文 WER 与说话人相似度"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":4,"references":[{"title":"Seed-TTS Eval","url":"https://github.com/BytedanceSpeech/seed-tts-eval"}]}},{"id":"text:audio:music","facet":{"shortLabel":"L2 助手级","items":["Music Generation Human Preference Benchmark：12 个模型、6,000 首歌、15,000 次盲选"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":2,"references":[{"title":"Music Generation Human Preference Benchmark","url":"https://arxiv.org/abs/2506.19085"}]}},{"id":"text:audio:effects","facet":{"shortLabel":"L2 助手级","items":["T2A-EpicBench：复杂多事件音效仍不稳定"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":2,"references":[{"title":"T2A-EpicBench","url":"https://aclanthology.org/2025.acl-long.1147/"}]}}]},{"id":"image:text","facet":{"shortLabel":"识图 / 还原","items":["读图识字","表格与 Word 还原","看图问答","图片定位"],"detail":"适合 OCR、复杂版式提取、看图问答和地理定位；关键数字与复杂表格仍需人工复核。","status":"ready","updatedAt":"2026-08-23"},"children":[{"id":"image:text:ocr","facet":{"shortLabel":"L3 熟练级","items":["OCRBench v2：10,000 条人工核验题、31 种场景；多数受测模型低于 50/100"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":3,"references":[{"title":"OCRBench v2","url":"https://arxiv.org/abs/2501.00321"}]}},{"id":"image:text:document","facet":{"shortLabel":"L2 助手级","items":["DREAM / DocRec1K：文档重建的 DSM、NED 与基线结果"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":2,"references":[{"title":"DREAM / DocRec1K","url":"https://arxiv.org/abs/2507.05805"}]}},{"id":"image:text:qa","facet":{"shortLabel":"L3 熟练级","items":["MMMU：11,500 题、30 种任务与人类对照"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":3,"references":[{"title":"MMMU","url":"https://openaccess.thecvf.com/content/CVPR2024/papers/Yue_MMMU_A_Massive_Multi-discipline_Multimodal_Understanding_and_Reasoning_Benchmark_for_CVPR_2024_paper.pdf"}]}},{"id":"image:text:geolocation","facet":{"shortLabel":"L2 助手级","items":["PIGEON：全球街景地理定位与距离误差，应用范围不等于通用图片定位"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":2,"references":[{"title":"PIGEON","url":"https://openaccess.thecvf.com/content/CVPR2024/html/Haas_PIGEON_Predicting_Image_Geolocations_CVPR_2024_paper.html"}]}}]},{"id":"image:image","facet":{"shortLabel":"修图 / 去背","items":["修图","局部重绘","风格转换","去背景"],"detail":"生成与编辑已经可以合并在一次对话里，但精细选区、透明边缘和批量一致性仍需检查。","status":"ready","updatedAt":"2026-08-23"},"children":[{"id":"image:image:retouch","facet":{"shortLabel":"L3 熟练级","items":["PPR10K：三位专家人像精修、人体区域与组间一致性"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":3,"references":[{"title":"PPR10K","url":"https://openaccess.thecvf.com/content/CVPR2021/html/Liang_PPR10K_A_Large-Scale_Portrait_Photo_Retouching_Dataset_With_Human-Region_Mask_CVPR_2021_paper.html"}]}},{"id":"image:image:redraw","facet":{"shortLabel":"L3 熟练级","items":["MagicBrush：一万余组人工编辑、多轮指令与模型结果"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":3,"references":[{"title":"MagicBrush","url":"https://proceedings.neurips.cc/paper_files/paper/2023/file/64008fa30cba9b4d1ab1bd3bd3d57d61-Paper-Datasets_and_Benchmarks.pdf"}]}},{"id":"image:image:style","facet":{"shortLabel":"L3 熟练级","items":["ArtFID：内容保留、风格质量与人类相关性实验"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":3,"references":[{"title":"ArtFID","url":"https://arxiv.org/abs/2207.12280"}]}},{"id":"image:image:cutout","facet":{"shortLabel":"L4 职业级","items":[],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":4,"references":[{"title":"DIS5K","url":"https://arxiv.org/abs/2203.03041"}]}}]},{"id":"image:video","facet":{"shortLabel":"图生视频","items":["图生视频","角色动画","同步声画生成"],"detail":"","status":"ready","updatedAt":"2026-08-23"},"children":[{"id":"image:video:generation","facet":{"shortLabel":"L2 助手级","items":["VBench I2V：主体一致性、动态、质量与人类对齐结果"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":2,"references":[{"title":"VBench I2V","url":"https://github.com/Vchitect/VBench"}]}},{"id":"image:video:character","facet":{"shortLabel":"L2 助手级","items":["AnimationBench：12 项动画原则、角色 / IP 保持与模型比较"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":2,"references":[{"title":"AnimationBench","url":"https://github.com/VideoVerses/AnimationBench"}]}},{"id":"image:video:sync-audio","facet":{"shortLabel":"L2 助手级","items":["VABench / AVGen-Bench：图片生成同步声画"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":2,"references":[{"title":"VABench","url":"https://openaccess.thecvf.com/content/CVPR2026/html/Hua_VABench_A_Comprehensive_Benchmark_for_Audio-Video_Generation_CVPR_2026_paper.html"},{"title":"AVGen-Bench","url":"https://github.com/microsoft/AVGen-Bench"}]}}]},{"id":"image:audio","facet":{"shortLabel":"讲图 / 配乐","items":["看图讲述","无障碍描述","图片配乐"],"detail":"","status":"ready","updatedAt":"2026-08-23"},"children":[{"id":"image:audio:narration","facet":{"shortLabel":"L3 熟练级","items":["Spoken Image Descriptions：Flickr8k 上的 BLEU、ROUGE-L 与端到端结果"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":3,"references":[{"title":"Synthesizing Spoken Descriptions of Images","url":"https://repository.tudelft.nl/record/uuid%3A914070cb-d4fe-4bad-8064-bc83154b895b"}]}},{"id":"image:audio:accessibility","facet":{"shortLabel":"L2 助手级","items":["Context Matters：54 组图像上下文、16 名盲人 / 低视力参与者与五维评分"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":2,"references":[{"title":"Context Matters","url":"https://elisakreiss.github.io/contextual-description-evaluation/paper/reflessmetrics.html"}]}},{"id":"image:audio:music","facet":{"shortLabel":"L1 演示级","items":["MeLFusion：图片与文本联合配乐，待主流产品实测"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":1,"references":[{"title":"MeLFusion","url":"https://openaccess.thecvf.com/content/CVPR2024/papers/Chowdhury_MeLFusion_Synthesizing_Music_from_Image_and_Language_Cues_using_Diffusion_CVPR_2024_paper.pdf"}]}}]},{"id":"video:text","facet":{"shortLabel":"字幕 / 总结","items":["字幕转写","内容总结","视频问答","课程笔记"],"detail":"适合处理 PPT 视频课程和长视频：先转写，再总结、提问或整理笔记；超长素材要留意上下文与上传限制。","status":"ready","updatedAt":"2026-08-23"},"children":[{"id":"video:text:subtitle","facet":{"shortLabel":"L3 熟练级","items":["SubER：同时计算转写、分段与时间轴错误"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":3,"references":[{"title":"SubER","url":"https://aclanthology.org/2022.iwslt-1.1/"}]}},{"id":"video:text:summary","facet":{"shortLabel":"L3 熟练级","items":["QEVA：800 份摘要 / 200 个视频，评估覆盖、事实性与时间连贯性"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":3,"references":[{"title":"QEVA","url":"https://aclanthology.org/2025.findings-emnlp.1340/"}]}},{"id":"video:text:qa","facet":{"shortLabel":"L2 助手级","items":["LVBench：超长视频问答与人类差距"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":2,"references":[{"title":"LVBench","url":"https://openaccess.thecvf.com/content/ICCV2025/html/Wang_LVBench_An_Extreme_Long_Video_Understanding_Benchmark_ICCV_2025_paper.html"}]}},{"id":"video:text:course","facet":{"shortLabel":"L2 助手级","items":["Video-MMLU：1,065 个课程视频，结构化笔记、VDCScore 与 QA"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":2,"references":[{"title":"Video-MMLU","url":"https://arxiv.org/abs/2504.14693"}]}}]},{"id":"video:image","facet":{"shortLabel":"抽帧 / 封面","items":["定点 / 等间隔抽帧","语义抽帧","封面选帧","缩略图重设计","分镜提取"],"detail":"","status":"ready","updatedAt":"2026-08-23"},"children":[{"id":"video:image:frame","facet":{"shortLabel":"L0 非 AI 任务","items":["FFmpeg 确定性抽帧"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":0,"references":[{"title":"FFmpeg","url":"https://ffmpeg.org/ffmpeg.html"}]}},{"id":"video:image:semantic-frame","facet":{"shortLabel":"L2 助手级","items":["TimeLens / KFS-Bench：语义时间定位与关键帧采样"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":2,"references":[{"title":"TimeLens-Bench","url":"https://github.com/TencentARC/TimeLens"},{"title":"KFS-Bench","url":"https://openaccess.thecvf.com/content/WACV2026/html/Li_KFS-Bench_Comprehensive_Evaluation_of_Key_Frame_Sampling_in_Long_Video_WACV_2026_paper.html"}]}},{"id":"video:image:cover-frame","facet":{"shortLabel":"L2 助手级","items":["缩略图选择基准：从视频中选择代表帧"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":2,"references":[{"title":"Micro-video Thumbnail Selection","url":"https://arxiv.org/abs/2112.14958"}]}},{"id":"video:image:thumbnail","facet":{"shortLabel":"L2 助手级","items":["缩略图重设计定为 L2；YouTube Test & Compare 只评估结果，不是 AI 重设计横评"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":2,"references":[{"title":"YouTube Test & Compare","url":"https://support.google.com/youtube/answer/13861714"}]}},{"id":"video:image:storyboard","facet":{"shortLabel":"L2 助手级","items":["Keyframe-Caption Summary：固定数量关键帧、说明、数据与两套基线"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":2,"references":[{"title":"Keyframe-Caption Summary","url":"https://aclanthology.org/2023.emnlp-main.457/"}]}}]},{"id":"video:video","facet":{"shortLabel":"素材补充","items":["素材补充","二创","剪辑","翻译","修复 / 超分"],"detail":"当前生成视频更适合作为素材补充与二创；翻译和成片仍需要剪辑流程兜底。","status":"ready","updatedAt":"2026-08-23"},"children":[{"id":"video:video:supplement","facet":{"shortLabel":"L1 演示级","items":["B-Script：专家标注的 B-roll 位置与内容推荐，仅支持窄场景演示"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":1,"references":[{"title":"B-Script","url":"https://arxiv.org/abs/1902.11216"}]}},{"id":"video:video:remix","facet":{"shortLabel":"L2 助手级","items":["MEDit-Bench：按不同叙事意图制作专业剪辑；模型仍落后专业编辑"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":2,"references":[{"title":"MEDit-Bench","url":"https://arxiv.org/abs/2607.25300"}]}},{"id":"video:video:edit","facet":{"shortLabel":"L2 助手级","items":["Anatomy of Video Editing / MEDit-Bench：素材组织、组装、时间对齐与人工偏好"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":2,"references":[{"title":"Anatomy of Video Editing","url":"https://arxiv.org/abs/2207.09812"},{"title":"MEDit-Bench","url":"https://arxiv.org/abs/2607.25300"}]}},{"id":"video:video:translation","facet":{"shortLabel":"L3 熟练级","items":["IWSLT Automatic Dubbing：翻译质量、自然度、等时性与声画同步"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":3,"references":[{"title":"IWSLT Automatic Dubbing","url":"https://aclanthology.org/2023.iwslt-1.1.pdf"}]}},{"id":"video:video:restore","facet":{"shortLabel":"L3 熟练级","items":["NTIRE 2026 KwaiVIR：真实 UGC 视频修复"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":3,"references":[{"title":"NTIRE 2026 Video Restoration Challenge","url":"https://arxiv.org/abs/2604.10551"}]}}]},{"id":"video:audio","facet":{"shortLabel":"提取 / 配音效","items":["音轨提取","重新配音","视频配音效 / Foley"],"detail":"","status":"ready","updatedAt":"2026-08-23"},"children":[{"id":"video:audio:extract","facet":{"shortLabel":"L0 非 AI 任务","items":["FFmpeg 确定性音轨提取"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":0,"references":[{"title":"FFmpeg","url":"https://ffmpeg.org/ffmpeg.html"}]}},{"id":"video:audio:dub","facet":{"shortLabel":"L3 熟练级","items":["VoiceCraft-Dub：WER、说话人相似度、MOS 与口型同步"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":3,"references":[{"title":"VoiceCraft-Dub","url":"https://www.openaccess.thecvf.com/content/ICCV2025/papers/Sung-Bin_VoiceCraft-Dub_Automated_Video_Dubbing_with_Neural_Codec_Language_Models_ICCV_2025_paper.pdf"}]}},{"id":"video:audio:effects","facet":{"shortLabel":"L2 助手级","items":["Video-Guided Foley：VGGSound、FAD、CLAP、AV-Sync 与 8,702 段视频"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":2,"references":[{"title":"Video-Guided Foley","url":"https://openaccess.thecvf.com/content/CVPR2025/papers/Chen_Video-Guided_Foley_Sound_Generation_with_Multimodal_Controls_CVPR_2025_paper.pdf"}]}}]},{"id":"audio:text","facet":{"shortLabel":"转写 / 纪要","items":["语音转写","会议纪要","歌词识别","说话人标注"],"detail":"","status":"ready","updatedAt":"2026-08-23"},"children":[{"id":"audio:text:transcribe","facet":{"shortLabel":"L4 职业级","items":["Open ASR Leaderboard：60 余个系统、11 个数据集、WER 与实时因子"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":4,"references":[{"title":"Open ASR Leaderboard","url":"https://arxiv.org/abs/2510.06961"}]}},{"id":"audio:text:minutes","facet":{"shortLabel":"L3 熟练级","items":["QMSum：232 场会议、1,808 个查询—摘要对"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":3,"references":[{"title":"QMSum","url":"https://aclanthology.org/2021.naacl-main.472/"}]}},{"id":"audio:text:lyrics","facet":{"shortLabel":"L2 助手级","items":["MIREX Automatic Lyrics Transcription：多数据集 WER / CER 与挑战结果"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":2,"references":[{"title":"MIREX Automatic Lyrics Transcription","url":"https://music-ir.org/mirex/wiki/2021%3AAutomatic_Lyrics_Transcription"}]}},{"id":"audio:text:diarization","facet":{"shortLabel":"L3 熟练级","items":["DIHARD III：多人录音中识别谁在何时说话"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":3,"references":[{"title":"DIHARD III","url":"https://dihardchallenge.github.io/dihard3/"}]}}]},{"id":"audio:image","facet":{"shortLabel":"专辑封面 / 频谱","items":["专辑封面","波形","频谱图"],"detail":"","status":"ready","updatedAt":"2026-08-23"},"children":[{"id":"audio:image:cover","facet":{"shortLabel":"L1 演示级","items":["Audio-guided Album Cover Art：单论文自测 FID 35.39、类别识别 40%，仅支持 L1 演示级"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":1,"references":[{"title":"Audio-guided Album Cover Art","url":"https://arxiv.org/html/2207.07162v1"}]}},{"id":"audio:image:waveform","facet":{"shortLabel":"L0 非 AI 任务","items":["FFmpeg showwaves / Audacity"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":0,"references":[{"title":"FFmpeg Filters Documentation","url":"https://www.ffmpeg.org/ffmpeg-filters.html"}]}},{"id":"audio:image:spectrum","facet":{"shortLabel":"L0 非 AI 任务","items":["FFmpeg showspectrum / Audacity"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":0,"references":[{"title":"FFmpeg Filters Documentation","url":"https://www.ffmpeg.org/ffmpeg-filters.html"}]}}]},{"id":"audio:video","facet":{"shortLabel":"MV / 口型","items":["MV","口型","数字人"],"detail":"","status":"ready","updatedAt":"2026-08-23"},"children":[{"id":"audio:video:mv","facet":{"shortLabel":"L1 演示级","items":["AutoMV-Bench：专家按音乐内容、技术、后期与艺术性 12 项评分"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":1,"references":[{"title":"AutoMV-Bench","url":"https://arxiv.org/abs/2512.12196"}]}},{"id":"audio:video:lip-sync","facet":{"shortLabel":"L3 熟练级","items":["THQA-10K：10,457 段数字人视频的主客观质量与同步指标"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":3,"references":[{"title":"THQA-10K","url":"https://openaccess.thecvf.com/content/ICCV2025/html/Zhou_Who_is_a_Better_Talker_Subjective_and_Objective_Quality_Assessment_ICCV_2025_paper.html"}]}},{"id":"audio:video:avatar","facet":{"shortLabel":"L2 助手级","items":["JWB-DH-V1：10,000 个身份、200 万视频的全身动作、语音与区域级评测"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":2,"references":[{"title":"JWB-DH-V1","url":"https://arxiv.org/abs/2507.20987"}]}}]},{"id":"audio:audio","facet":{"shortLabel":"降噪 / 分轨","items":["降噪","分轨","变声","续作","语音翻译"],"detail":"","status":"ready","updatedAt":"2026-08-23"},"children":[{"id":"audio:audio:denoise","facet":{"shortLabel":"L4 职业级","items":["DNS Challenge：真实噪声与主观听测"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":4,"references":[{"title":"DNS Challenge","url":"https://www.microsoft.com/en-us/research/academic-program/deep-noise-suppression-challenge-icassp-2023/results/"}]}},{"id":"audio:audio:separate","facet":{"shortLabel":"L4 职业级","items":["MUSDB18 / SiSEC：复杂混音仍有串音与伪影"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":4,"references":[{"title":"MUSDB18 / SiSEC","url":"https://sigsep.github.io/datasets/musdb.html"}]}},{"id":"audio:audio:voice","facet":{"shortLabel":"L3 熟练级","items":["Voice Conversion Challenge：跨语言、情绪与唱声泛化仍有限"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":3,"references":[{"title":"Voice Conversion Challenge","url":"https://www.vc-challenge.org/"}]}},{"id":"audio:audio:continue","facet":{"shortLabel":"L1 演示级","items":["Suno 续作功能已上线，但没有公开可核验的专项测评"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":1,"references":[{"title":"Suno Extend 功能说明","url":"https://help.suno.com/en/articles/2409601"},{"title":"Suno Song Editor 功能说明","url":"https://help.suno.com/en/articles/6141505"}]}},{"id":"audio:audio:translation","facet":{"shortLabel":"L3 熟练级","items":["Seed LiveInterpret 2.0：官方中英同传人评 74.8 / 100、约 3 秒延迟；尚缺独立横评"],"detail":"","status":"ready","updatedAt":"2026-08-23","capabilityLevel":3,"references":[{"title":"Seed LiveInterpret 2.0 / 豆包同声传译 2.0","url":"https://seed.bytedance.com/en/blog/seed-liveinterpret-2-0-released-an-end-to-end-simultaneous-interpretation-model-featuring-ultra-high-accuracy-close-to-human-interpreters-low-latency-of-3-seconds-and-real-time-voice-cloning"},{"title":"SeamlessM4T","url":"https://www.nature.com/articles/s41586-024-08359-z"}]}}]}]}},"meta":{"version":"2","updatedAt":"2026-08-13T02:13:52.784Z"}}