{"text": "相信你也被GPT6操作Blender建模的視頻給刷屏了吧。而本期我們挑戰點不一樣的,用本地部署的小模型操作Blender,復刻GPT6的效果,並分享完整的過程。為了避免浪費大家的時間,我這裡先給大家看成果,以便大家決定要不要繼續往下看。我這裡使用的是千萬3.820GB的iQ4量化板。操作Blender製作了一個三線二的效果,像這自行車、車框、輪骨都製作了蠻精細的,包括動漫風格的描邊,它也正確生成了。旁邊飲料機中還有帶金屬質感的一拉罐,商店另一側的供貨板生成的也挺詳細的。以及商店內的貨架和貨物都進行了詳細的生成。這對於一個只有27B參數,還是一個iQ4量化板的小模型來說,真的挺不容易的。你之所以選擇iQ4的版本,是為了照顧一下顯存不足的小夥伴,讓更多的人可以跟著一起做。當然你選擇線上的API也是完全OK的,選擇便宜的具有性價比的模型即可。我們重點是講解思路和處理問題的方法,而不是必須選擇某個模型才能實現這樣的效果。當你掌握了這一套方法,你將不必被某一個模型能力縮左右,讓你成為一個真正的Webcoder。我是ZZ,就讓我們正式開始吧。ZZZ,這次是GitHub的GitHub,並獲得不錯的效果,也是用的BlendMCP。我這裡直接在OpenCode中,讓其在GitHub中下載BlendMCP並安裝。這裡可以把Blend的目錄一併提供。完成後我們來聊聊首輪提示詞。在上一期製作FPS的視頻中,我們對生成的步驟進行了拆分,僅在首輪提示詞中明確了核心玩法關鍵需求。在確認生成結果無誤後,才繼續後續的多個步驟,並逐步完成了整個遊戲。我們請就想和大家好好聊一下步驟拆分的問題。其實步驟拆分最大的決定因素是模型自身。如果模型過於孱弱,你不拆分,最終的結果就是一坨。而模型越強,則越有可能給你一期合成。其次就是提示詞之間的非代碼層面是否完全可接偶。這句話有點抽象。我給大家舉兩個例子,大家就懂了。第一個例子還是剛才的FPS遊戲。我們把核心玩法和場景拆成兩步。首輪提示詞只定義核心玩法,場景後製生成。哪怕只有一個基礎平面,射擊敵人的核心機制依舊是可以正常運行的。這種就非常適合分布實現。第二個例子是我此前視頻中製作的殺核破壞網遊戲。點擊破壞場景中物體這個交互,和場景本身深度保定。如果先單獨生成場景,產出的模型結構很可能不適合做破壞的邏輯。後續的修補成本極高。這類任務就更適合一次性生成。就算手輪效果不夠完美,後續迭代修改也更加順暢。而像我們本期不涉及玩法,只是單純的用Blend建模。我們的場景中存在很多不同的元素,但同時又需要風格統一且布局合理。因此更適合一口氣告訴AI,讓其進行合理發揮,以減少後續的修補工作。我們來看一下詳細的手輪提示詞。第一段相當於一個總體,整體說明了一下我想要製作的是一個什麼樣的場景以及什麼風格的。然後我強調生成了agent.md文件,因為我希望AI能不安我一口氣生成。中途不要問我。agent.md文件可以很好的在長時間運行的任務中框定agent的行為。接著我補充了一下Blend的安裝位置。這是為了避免agent尋找安裝位置而浪費時間。然後是架構方面的約束,我們希望每個模型都是單獨的文件,最後再進行組裝。同時要求製作詳細的步驟清單並依次執行。接著我們要求其質量優先。美術風格方面,我們希望它色彩如何治愈,輪廓清晰乾淨。場景布局希望它動線合理,尺寸緊湊。接著是戶外資產和便利店內部的相關元素。我們把名稱都給羅列清楚,以增加場景的豐富度。渲染標準這裡我們強調出區分材質,並且渲染陰影,明暗清晰。在效果這裡我們希望增加櫻花飄落的動畫。最後再次強調了一下輸出的硬性約束。所以我們就直接讓它執行。操作Blend鍵模式非常消耗尚氧文的一個事情。所以在運行的過程中,一旦遇到超過尚氧文的情況,不要緊張。我們直接先開一個對話,然後繼續執行即可。等待一會後,我們就可以看到生存效果了。第一次能射成成這樣,我覺得已經很厲害了。然後我們來修改一下其中的一些問題。比如說這個櫻花樹的樹葉它太分散了,包括樹枝的朝向不太正確。完成後,我們打開渲染。然後可以看到它這個沒有陰影啊。這個其實是它材質的問題。我們讓它修改一下。可以看到AI居然先給我們做了一個測試。真的是有夠嚴謹的。可以看到現在這個光影效果就有了。不過啊,三軒二的那種效果還是不夠強烈。尤其是沒有描邊。這裡我們再讓它修改一下。好,這下看起來顏色風格什麼的就對了嘛。我發現自行系的前輪胎的內胎的旋轉方向有點問題。這時候就發揮Blender的優勢了。我們直接在Blender調整參數來修改啊。我覺得便電店旁邊這個地方有點太空曠了。然後我讓它先登一個小汽車。然後加上一個廣告牌。本地小模型操作Blender建模的效果和GPD6相比,你覺得有多少差距呢?想用千聞3.8的漆皮建模,其專業領域還是在3.js上。大家看,這是我用同樣的提示詞,只不過使用的是3.js生成的。是不是效果提升一大截呢?實際上,如果你會用Blender,你可以簡單調整,就可以獲得更好的渲染效果。也就是說,Blender其實上限更高。但是如果你只是在網頁上實現一些程序化的生成的模型。那麼3.js仍然是優先選擇。因為直接生成3.js,比起用AI來操作Blender的速度要快速唄。如果你想了解如何使用3.js建模,為你的網站搭接一些3D元素或動向。關注我吧。後續我將分享如何使用3.js制作這樣的場景。感謝大家看到這裡。我是ZZ,期待下次相遇。", "segments": [{"id": 0, "seek": 0, "start": 0.0, "end": 13.0, "text": "相信你也被GPT6操作Blender建模的視頻給刷屏了吧。而本期我們挑戰點不一樣的,用本地部署的小模型操作Blender,復刻GPT6的效果,並分享完整的過程。", "tokens": [50365, 35428, 2166, 6404, 23238, 38, 47, 51, 21, 13167, 235, 11914, 14520, 3216, 34157, 41908, 1546, 27333, 29940, 17798, 2437, 115, 9636, 237, 37091, 1543, 11070, 8802, 16786, 5884, 49067, 23209, 8216, 1960, 22977, 1546, 11, 9254, 8802, 10928, 13470, 16469, 110, 1546, 7322, 41908, 39823, 13167, 235, 11914, 14520, 3216, 11, 2172, 102, 45500, 38, 47, 51, 21, 1546, 43076, 9319, 11, 35934, 30855, 14128, 27662, 1546, 8816, 29649, 1543, 51015], "temperature": 0, "avg_logprob": -0.1624162397311844, "compression_ratio": 1.1851851851851851, "no_speech_prob": 8.379638996580674e-12}, {"id": 1, "seek": 0, "start": 13.0, "end": 23.0, "text": "為了避免浪費大家的時間,我這裡先給大家看成果,以便大家決定要不要繼續往下看。我這裡使用的是千萬3.820GB的iQ4量化板。", "tokens": [51015, 6344, 2289, 3330, 123, 2347, 235, 11038, 103, 50071, 6868, 1546, 20788, 11, 1654, 18771, 10108, 17798, 6868, 4200, 11336, 9319, 11, 3588, 27364, 6868, 33540, 12088, 4275, 11962, 38459, 29510, 4438, 4200, 1543, 1654, 18771, 22982, 9254, 24620, 20787, 14099, 18, 13, 23, 2009, 8769, 1546, 72, 48, 19, 26748, 23756, 43664, 1543, 51515], "temperature": 0, "avg_logprob": -0.1624162397311844, "compression_ratio": 1.1851851851851851, "no_speech_prob": 8.379638996580674e-12}, {"id": 2, "seek": 2300, "start": 23.0, "end": 37.0, "text": "操作Blender製作了一個三線二的效果,像這自行車、車框、輪骨都製作了蠻精細的,包括動漫風格的描邊,它也正確生成了。旁邊飲料機中還有帶金屬質感的一拉罐,商店另一側的供貨板生成的也挺詳細的。", "tokens": [50365, 13167, 235, 11914, 14520, 3216, 8083, 121, 11914, 2289, 8990, 10960, 31699, 11217, 1546, 43076, 9319, 11, 12760, 2664, 9722, 8082, 25941, 1231, 25941, 21416, 228, 1231, 12290, 103, 165, 16838, 7182, 8083, 121, 11914, 2289, 164, 254, 119, 34910, 41588, 1546, 11, 41828, 12572, 14065, 104, 22713, 30921, 1546, 11673, 237, 14901, 11, 11284, 6404, 15789, 24293, 8244, 11336, 2289, 1543, 4479, 223, 14901, 11808, 110, 33404, 17543, 5975, 15569, 32379, 19117, 9636, 105, 43450, 9709, 1546, 2257, 29717, 16469, 238, 11, 45581, 22469, 22762, 2257, 7437, 112, 1546, 3254, 249, 11561, 101, 43664, 8244, 11336, 1546, 6404, 41046, 6716, 111, 41588, 1546, 1543, 51065], "temperature": 0, "avg_logprob": -0.18012265461247143, "compression_ratio": 1.284848484848485, "no_speech_prob": 1.0787387476340893e-11}, {"id": 3, "seek": 2300, "start": 37.0, "end": 47.0, "text": "以及商店內的貨架和貨物都進行了詳細的生成。這對於一個只有27B參數,還是一個iQ4量化板的小模型來說,真的挺不容易的。", "tokens": [51065, 40282, 45581, 22469, 28472, 1546, 11561, 101, 7360, 114, 12565, 11561, 101, 23516, 7182, 18214, 46606, 6716, 111, 41588, 1546, 8244, 11336, 1543, 2664, 2855, 19488, 8990, 35244, 10076, 33, 47375, 30622, 11, 25583, 8990, 72, 48, 19, 26748, 23756, 43664, 1546, 7322, 41908, 39823, 36979, 11, 8034, 41046, 1960, 49212, 1546, 1543, 51565], "temperature": 0, "avg_logprob": -0.18012265461247143, "compression_ratio": 1.284848484848485, "no_speech_prob": 1.0787387476340893e-11}, {"id": 4, "seek": 4700, "start": 47.0, "end": 58.0, "text": "你之所以選擇iQ4的版本,是為了照顧一下顯存不足的小夥伴,讓更多的人可以跟著一起做。當然你選擇線上的API也是完全OK的,選擇便宜的具有性價比的模型即可。", "tokens": [50365, 2166, 9574, 7239, 21625, 13167, 229, 72, 48, 19, 1546, 42096, 8802, 11, 1541, 6344, 2289, 32150, 10178, 100, 8861, 46605, 39781, 1960, 37236, 1546, 7322, 1787, 98, 7384, 112, 11, 21195, 19002, 6392, 29979, 6723, 9678, 19382, 29567, 10907, 1543, 21707, 2166, 21625, 13167, 229, 31699, 5708, 1546, 4715, 40, 22021, 37100, 9443, 1546, 11, 21625, 13167, 229, 27364, 2415, 250, 1546, 39806, 2412, 21686, 45441, 11706, 1546, 41908, 39823, 39127, 4429, 1543, 50915], "temperature": 0, "avg_logprob": -0.09396708493976924, "compression_ratio": 1.298913043478261, "no_speech_prob": 1.7164967364147188e-11}, {"id": 5, "seek": 4700, "start": 58.0, "end": 70.0, "text": "我們重點是講解思路和處理問題的方法,而不是必須選擇某個模型才能實現這樣的效果。當你掌握了這一套方法,你將不必被某一個模型能力縮左右,讓你成為一個真正的Webcoder。", "tokens": [50915, 5884, 12624, 8216, 1541, 11932, 17278, 8870, 24658, 12565, 39289, 13876, 17197, 1546, 9249, 11148, 11, 11070, 7296, 28531, 8313, 230, 21625, 13167, 229, 17238, 238, 3338, 41908, 39823, 18888, 8225, 10376, 9581, 39425, 43076, 9319, 1543, 13118, 2166, 6900, 234, 11673, 94, 2289, 32260, 1881, 245, 9249, 11148, 11, 2166, 34168, 1960, 28531, 23238, 17238, 238, 8990, 41908, 39823, 8225, 13486, 21030, 106, 29457, 11, 21195, 2166, 11336, 6344, 8990, 6303, 15789, 1546, 4360, 65, 66, 19866, 1543, 51515], "temperature": 0, "avg_logprob": -0.09396708493976924, "compression_ratio": 1.298913043478261, "no_speech_prob": 1.7164967364147188e-11}, {"id": 6, "seek": 4700, "start": 70.0, "end": 73.0, "text": "我是ZZ,就讓我們正式開始吧。", "tokens": [51515, 15914, 57, 57, 11, 3111, 21195, 5884, 15789, 27584, 21017, 6062, 1543, 51665], "temperature": 0, "avg_logprob": -0.09396708493976924, "compression_ratio": 1.298913043478261, "no_speech_prob": 1.7164967364147188e-11}, {"id": 7, "seek": 7300, "start": 73.0, "end": 82.0, "text": "ZZZ,這次是GitHub的GitHub,並獲得不錯的效果,也是用的BlendMCP。我這裡直接在OpenCode中,讓其在GitHub中下載BlendMCP並安裝。", "tokens": [50365, 57, 57, 57, 11, 2664, 9487, 1541, 38, 270, 21150, 1546, 38, 270, 21150, 11, 35934, 29057, 110, 5916, 30967, 1546, 43076, 9319, 11, 22021, 9254, 1546, 14520, 521, 44, 20049, 1543, 1654, 18771, 43297, 3581, 45569, 34, 1429, 5975, 11, 21195, 9572, 3581, 38, 270, 21150, 5975, 4438, 12290, 231, 14520, 521, 44, 20049, 35934, 16206, 44453, 1543, 50815], "temperature": 0, "avg_logprob": -0.2518165397644043, "compression_ratio": 1.3358585858585859, "no_speech_prob": 1.494506601806833e-11}, {"id": 8, "seek": 7300, "start": 82.0, "end": 94.0, "text": "這裡可以把Blend的目錄一併提供。完成後我們來聊聊首輪提示詞。在上一期製作FPS的視頻中,我們對生成的步驟進行了拆分,僅在首輪提示詞中明確了核心玩法關鍵需求。", "tokens": [50815, 18771, 6723, 16075, 14520, 521, 1546, 11386, 9567, 226, 2257, 1593, 113, 20949, 3254, 249, 1543, 41509, 5661, 5884, 3763, 40096, 40096, 25444, 12290, 103, 20949, 25696, 6716, 252, 1543, 3581, 5708, 2257, 16786, 8083, 121, 11914, 37, 6273, 1546, 27333, 29940, 5975, 11, 5884, 2855, 8244, 11336, 1546, 31429, 24023, 253, 18214, 46606, 6852, 228, 6627, 11, 9502, 227, 3581, 25444, 12290, 103, 20949, 25696, 6716, 252, 5975, 11100, 24293, 2289, 8987, 116, 7945, 19912, 11148, 14899, 41668, 113, 32535, 32718, 1543, 51415], "temperature": 0, "avg_logprob": -0.2518165397644043, "compression_ratio": 1.3358585858585859, "no_speech_prob": 1.494506601806833e-11}, {"id": 9, "seek": 7300, "start": 94.0, "end": 102.0, "text": "在確認生成結果無誤後,才繼續後續的多個步驟,並逐步完成了整個遊戲。我們請就想和大家好好聊一下步驟拆分的問題。", "tokens": [51415, 3581, 24293, 22041, 8244, 11336, 35181, 16976, 3549, 97, 5661, 11, 18888, 38459, 5661, 29583, 1546, 6392, 3338, 31429, 24023, 253, 11, 35934, 2215, 238, 31429, 41509, 2289, 27662, 3338, 43046, 1543, 5884, 16302, 3111, 7093, 12565, 15248, 2131, 40096, 8861, 31429, 24023, 253, 6852, 228, 6627, 1546, 17197, 1543, 51815], "temperature": 0, "avg_logprob": -0.2518165397644043, "compression_ratio": 1.3358585858585859, "no_speech_prob": 1.494506601806833e-11}, {"id": 10, "seek": 10200, "start": 102.0, "end": 111.0, "text": "其實步驟拆分最大的決定因素是模型自身。如果模型過於孱弱,你不拆分,最終的結果就是一坨。而模型越強,則越有可能給你一期合成。", "tokens": [50365, 14139, 31429, 24023, 253, 6852, 228, 6627, 8661, 39156, 33540, 12088, 8698, 37204, 1541, 41908, 39823, 9722, 19847, 1543, 13119, 41908, 39823, 8816, 19488, 8052, 109, 5702, 109, 11, 2166, 1960, 6852, 228, 6627, 11, 8661, 29371, 1546, 35181, 5620, 2257, 14872, 101, 1543, 11070, 41908, 39823, 25761, 22752, 11, 46225, 25761, 2412, 16657, 17798, 2166, 2257, 16786, 14245, 11336, 1543, 50815], "temperature": 0, "avg_logprob": -0.13764441864831106, "compression_ratio": 1.2741935483870968, "no_speech_prob": 1.4877007611935333e-11}, {"id": 11, "seek": 10200, "start": 111.0, "end": 118.0, "text": "其次就是提示詞之間的非代碼層面是否完全可接偶。這句話有點抽象。我給大家舉兩個例子,大家就懂了。", "tokens": [50815, 9572, 9487, 5620, 20949, 25696, 6716, 252, 9574, 11016, 1546, 12107, 19105, 16337, 120, 9636, 97, 8833, 1541, 45334, 37100, 4429, 14468, 7437, 114, 1543, 2664, 34592, 11103, 42440, 46022, 45007, 1543, 1654, 17798, 6868, 42828, 34623, 17797, 7626, 11, 6868, 3111, 29624, 2289, 1543, 51165], "temperature": 0, "avg_logprob": -0.13764441864831106, "compression_ratio": 1.2741935483870968, "no_speech_prob": 1.4877007611935333e-11}, {"id": 12, "seek": 11800, "start": 118.0, "end": 126.0, "text": "第一個例子還是剛才的FPS遊戲。我們把核心玩法和場景拆成兩步。首輪提示詞只定義核心玩法,場景後製生成。", "tokens": [50365, 40760, 17797, 7626, 25583, 16940, 18888, 1546, 37, 6273, 43046, 1543, 5884, 16075, 8987, 116, 7945, 19912, 11148, 12565, 21446, 50218, 6852, 228, 11336, 16313, 31429, 1543, 25444, 12290, 103, 20949, 25696, 6716, 252, 14003, 12088, 37430, 8987, 116, 7945, 19912, 11148, 11, 21446, 50218, 5661, 8083, 121, 8244, 11336, 1543, 50765], "temperature": 0, "avg_logprob": -0.07992086978937617, "compression_ratio": 1.3475409836065573, "no_speech_prob": 1.6394099619509994e-11}, {"id": 13, "seek": 11800, "start": 126.0, "end": 133.0, "text": "哪怕只有一個基礎平面,射擊敵人的核心機制依舊是可以正常運行的。這種就非常適合分布實現。", "tokens": [50765, 17028, 21164, 35244, 8990, 26008, 14971, 236, 16716, 8833, 11, 1530, 226, 37813, 7017, 113, 4035, 1546, 8987, 116, 7945, 17543, 25491, 3254, 251, 10256, 232, 1541, 6723, 15789, 11279, 32772, 8082, 1546, 1543, 33005, 3111, 14392, 3330, 102, 14245, 6627, 34688, 10376, 9581, 1543, 51115], "temperature": 0, "avg_logprob": -0.07992086978937617, "compression_ratio": 1.3475409836065573, "no_speech_prob": 1.6394099619509994e-11}, {"id": 14, "seek": 11800, "start": 133.0, "end": 141.0, "text": "第二個例子是我此前視頻中製作的殺核破壞網遊戲。點擊破壞場景中物體這個交互,和場景本身深度保定。", "tokens": [51115, 49085, 17797, 7626, 42614, 17947, 8945, 27333, 29940, 5975, 8083, 121, 11914, 1546, 45528, 8987, 116, 39560, 12829, 252, 26352, 43046, 1543, 8216, 37813, 39560, 12829, 252, 21446, 50218, 5975, 23516, 23987, 6287, 28455, 1369, 240, 11, 12565, 21446, 50218, 8802, 19847, 24043, 13127, 24302, 12088, 1543, 51515], "temperature": 0, "avg_logprob": -0.07992086978937617, "compression_ratio": 1.3475409836065573, "no_speech_prob": 1.6394099619509994e-11}, {"id": 15, "seek": 14100, "start": 141.0, "end": 150.0, "text": "如果先單獨生成場景,產出的模型結構很可能不適合做破壞的邏輯。後續的修補成本極高。這類任務就更適合一次性生成。", "tokens": [50365, 13119, 10108, 24227, 163, 18304, 8244, 11336, 21446, 50218, 11, 33299, 7781, 1546, 41908, 39823, 17144, 43362, 4563, 16657, 1960, 3330, 102, 14245, 10907, 39560, 12829, 252, 1546, 3023, 237, 12290, 107, 1543, 5661, 29583, 1546, 7792, 106, 8083, 250, 11336, 8802, 13387, 113, 12979, 1543, 2664, 40092, 26443, 39517, 3111, 19002, 3330, 102, 14245, 27505, 21686, 8244, 11336, 1543, 50815], "temperature": 0, "avg_logprob": -0.0768395260552675, "compression_ratio": 1.3240997229916898, "no_speech_prob": 1.8284956881942094e-11}, {"id": 16, "seek": 14100, "start": 150.0, "end": 162.0, "text": "就算手輪效果不夠完美,後續迭代修改也更加順暢。而像我們本期不涉及玩法,只是單純的用Blend建模。我們的場景中存在很多不同的元素,但同時又需要風格統一且布局合理。", "tokens": [50815, 3111, 19497, 11389, 12290, 103, 43076, 9319, 1960, 31649, 14128, 9175, 11, 5661, 29583, 3316, 255, 19105, 7792, 106, 34490, 6404, 19002, 9990, 8313, 228, 23826, 95, 1543, 11070, 12760, 5884, 8802, 16786, 1960, 35681, 231, 25703, 19912, 11148, 11, 36859, 24227, 7732, 242, 1546, 9254, 14520, 521, 34157, 41908, 1543, 33641, 21446, 50218, 5975, 39781, 3581, 20778, 47123, 1546, 14812, 37204, 11, 8395, 13089, 6611, 17047, 35748, 22713, 30921, 33725, 2257, 20334, 34688, 34703, 14245, 13876, 1543, 51415], "temperature": 0, "avg_logprob": -0.0768395260552675, "compression_ratio": 1.3240997229916898, "no_speech_prob": 1.8284956881942094e-11}, {"id": 17, "seek": 14100, "start": 162.0, "end": 167.0, "text": "因此更適合一口氣告訴AI,讓其進行合理發揮,以減少後續的修補工作。", "tokens": [51415, 8698, 17947, 19002, 3330, 102, 14245, 2257, 18144, 24090, 31080, 48698, 11, 21195, 9572, 18214, 8082, 14245, 13876, 14637, 11673, 106, 11, 3588, 9592, 249, 15686, 5661, 29583, 1546, 7792, 106, 8083, 250, 41315, 1543, 51665], "temperature": 0, "avg_logprob": -0.0768395260552675, "compression_ratio": 1.3240997229916898, "no_speech_prob": 1.8284956881942094e-11}, {"id": 18, "seek": 16700, "start": 167.0, "end": 175.0, "text": "我們來看一下詳細的手輪提示詞。第一段相當於一個總體,整體說明了一下我想要製作的是一個什麼樣的場景以及什麼風格的。", "tokens": [50365, 5884, 3763, 28324, 6716, 111, 41588, 1546, 11389, 12290, 103, 20949, 25696, 6716, 252, 1543, 18049, 28427, 15106, 13118, 19488, 8990, 26575, 23987, 11, 27662, 23987, 4622, 11100, 2289, 8861, 25246, 4275, 8083, 121, 11914, 24620, 8990, 7598, 6180, 1546, 21446, 50218, 40282, 7598, 22713, 30921, 1546, 1543, 50765], "temperature": 0, "avg_logprob": -0.10935633437438581, "compression_ratio": 1.3440233236151604, "no_speech_prob": 2.212299787807126e-11}, {"id": 19, "seek": 16700, "start": 175.0, "end": 187.0, "text": "然後我強調生成了agent.md文件,因為我希望AI能不安我一口氣生成。中途不要問我。agent.md文件可以很好的在長時間運行的任務中框定agent的行為。", "tokens": [50765, 10213, 1654, 22752, 32943, 8244, 11336, 2289, 559, 317, 13, 76, 67, 17174, 20485, 11, 11471, 1654, 29955, 48698, 8225, 1960, 16206, 1654, 2257, 18144, 24090, 8244, 11336, 1543, 5975, 2215, 242, 11962, 11361, 1654, 1543, 559, 317, 13, 76, 67, 17174, 20485, 6723, 4563, 20715, 3581, 15353, 20788, 32772, 8082, 1546, 26443, 39517, 5975, 21416, 228, 12088, 559, 317, 1546, 8082, 6344, 1543, 51365], "temperature": 0, "avg_logprob": -0.10935633437438581, "compression_ratio": 1.3440233236151604, "no_speech_prob": 2.212299787807126e-11}, {"id": 20, "seek": 16700, "start": 187.0, "end": 193.0, "text": "接著我補充了一下Blend的安裝位置。這是為了避免agent尋找安裝位置而浪費時間。", "tokens": [51365, 14468, 19382, 1654, 8083, 250, 2347, 227, 2289, 8861, 14520, 521, 1546, 16206, 44453, 11160, 34719, 1543, 16600, 6344, 2289, 3330, 123, 2347, 235, 559, 317, 1530, 233, 25085, 16206, 44453, 11160, 34719, 11070, 11038, 103, 50071, 20788, 1543, 51665], "temperature": 0, "avg_logprob": -0.10935633437438581, "compression_ratio": 1.3440233236151604, "no_speech_prob": 2.212299787807126e-11}, {"id": 21, "seek": 19300, "start": 193.0, "end": 199.0, "text": "然後是架構方面的約束,我們希望每個模型都是單獨的文件,最後再進行組裝。", "tokens": [50365, 10213, 1541, 7360, 114, 43362, 39604, 1546, 30889, 44320, 11, 5884, 29955, 23664, 3338, 41908, 39823, 22796, 24227, 163, 18304, 1546, 17174, 20485, 11, 20578, 8623, 18214, 8082, 30052, 44453, 1543, 50665], "temperature": 0, "avg_logprob": -0.08551717297784213, "compression_ratio": 1.3031358885017421, "no_speech_prob": 1.71075376087515e-11}, {"id": 22, "seek": 19300, "start": 199.0, "end": 205.0, "text": "同時要求製作詳細的步驟清單並依次執行。接著我們要求其質量優先。", "tokens": [50665, 13089, 6611, 4275, 32718, 8083, 121, 11914, 6716, 111, 41588, 1546, 31429, 24023, 253, 21784, 24227, 35934, 3254, 251, 9487, 16262, 115, 8082, 1543, 14468, 19382, 5884, 4275, 32718, 9572, 43450, 26748, 12827, 103, 10108, 1543, 50965], "temperature": 0, "avg_logprob": -0.08551717297784213, "compression_ratio": 1.3031358885017421, "no_speech_prob": 1.71075376087515e-11}, {"id": 23, "seek": 19300, "start": 205.0, "end": 209.0, "text": "美術風格方面,我們希望它色彩如何治愈,輪廓清晰乾淨。", "tokens": [50965, 9175, 9890, 241, 22713, 30921, 39604, 11, 5884, 29955, 11284, 17673, 7391, 102, 43526, 30118, 4820, 230, 11, 12290, 103, 17527, 241, 21784, 5094, 108, 39698, 14176, 101, 1543, 51165], "temperature": 0, "avg_logprob": -0.08551717297784213, "compression_ratio": 1.3031358885017421, "no_speech_prob": 1.71075376087515e-11}, {"id": 24, "seek": 19300, "start": 209.0, "end": 215.0, "text": "場景布局希望它動線合理,尺寸緊湊。接著是戶外資產和便利店內部的相關元素。", "tokens": [51165, 21446, 50218, 34688, 34703, 29955, 11284, 12572, 31699, 14245, 13876, 11, 1530, 118, 4510, 116, 30172, 33744, 232, 1543, 14468, 19382, 1541, 1486, 114, 12022, 35380, 33299, 12565, 27364, 23700, 22469, 28472, 13470, 1546, 15106, 14899, 14812, 37204, 1543, 51465], "temperature": 0, "avg_logprob": -0.08551717297784213, "compression_ratio": 1.3031358885017421, "no_speech_prob": 1.71075376087515e-11}, {"id": 25, "seek": 21500, "start": 215.0, "end": 225.0, "text": "我們把名稱都給羅列清楚,以增加場景的豐富度。渲染標準這裡我們強調出區分材質,並且渲染陰影,明暗清晰。", "tokens": [50365, 5884, 16075, 15940, 49606, 7182, 17798, 36004, 43338, 38433, 11, 3588, 24228, 252, 9990, 21446, 50218, 1546, 19517, 238, 47564, 13127, 1543, 9592, 110, 46568, 46299, 28220, 18771, 5884, 22752, 32943, 7781, 44526, 6627, 4422, 238, 43450, 11, 35934, 20334, 9592, 110, 46568, 8842, 108, 16820, 11, 11100, 23826, 245, 21784, 5094, 108, 1543, 50865], "temperature": 0, "avg_logprob": -0.14242118552879052, "compression_ratio": 1.2527075812274369, "no_speech_prob": 1.502323092317237e-11}, {"id": 26, "seek": 21500, "start": 225.0, "end": 228.0, "text": "在效果這裡我們希望增加櫻花飄落的動畫。", "tokens": [50865, 3581, 43076, 9319, 18771, 5884, 29955, 24228, 252, 9990, 162, 104, 119, 20127, 11808, 226, 30848, 1546, 12572, 43822, 1543, 51015], "temperature": 0, "avg_logprob": -0.14242118552879052, "compression_ratio": 1.2527075812274369, "no_speech_prob": 1.502323092317237e-11}, {"id": 27, "seek": 21500, "start": 228.0, "end": 232.0, "text": "最後再次強調了一下輸出的硬性約束。", "tokens": [51015, 20578, 8623, 9487, 22752, 32943, 2289, 8861, 12290, 116, 7781, 1546, 38114, 105, 21686, 30889, 44320, 1543, 51215], "temperature": 0, "avg_logprob": -0.14242118552879052, "compression_ratio": 1.2527075812274369, "no_speech_prob": 1.502323092317237e-11}, {"id": 28, "seek": 21500, "start": 232.0, "end": 234.0, "text": "所以我們就直接讓它執行。", "tokens": [51215, 7239, 5884, 3111, 43297, 21195, 11284, 16262, 115, 8082, 1543, 51315], "temperature": 0, "avg_logprob": -0.14242118552879052, "compression_ratio": 1.2527075812274369, "no_speech_prob": 1.502323092317237e-11}, {"id": 29, "seek": 21500, "start": 234.0, "end": 237.0, "text": "操作Blend鍵模式非常消耗尚氧文的一個事情。", "tokens": [51315, 13167, 235, 11914, 14520, 521, 41668, 113, 41908, 27584, 14392, 28837, 4450, 245, 1530, 248, 6610, 100, 17174, 1546, 8990, 24675, 1543, 51465], "temperature": 0, "avg_logprob": -0.14242118552879052, "compression_ratio": 1.2527075812274369, "no_speech_prob": 1.502323092317237e-11}, {"id": 30, "seek": 23700, "start": 237.0, "end": 241.0, "text": "所以在運行的過程中,一旦遇到超過尚氧文的情況,不要緊張。", "tokens": [50365, 7239, 3581, 32772, 8082, 1546, 8816, 29649, 5975, 11, 2257, 4479, 99, 3330, 229, 4511, 19869, 8816, 1530, 248, 6610, 100, 17174, 1546, 39391, 11, 11962, 30172, 18889, 1543, 50565], "temperature": 0, "avg_logprob": -0.11965538340912768, "compression_ratio": 1.4297994269340975, "no_speech_prob": 1.5087918761591546e-11}, {"id": 31, "seek": 23700, "start": 241.0, "end": 245.0, "text": "我們直接先開一個對話,然後繼續執行即可。", "tokens": [50565, 5884, 43297, 10108, 8949, 8990, 2855, 11103, 11, 10213, 38459, 16262, 115, 8082, 39127, 4429, 1543, 50765], "temperature": 0, "avg_logprob": -0.11965538340912768, "compression_ratio": 1.4297994269340975, "no_speech_prob": 1.5087918761591546e-11}, {"id": 32, "seek": 23700, "start": 245.0, "end": 248.0, "text": "等待一會後,我們就可以看到生存效果了。", "tokens": [50765, 10187, 18390, 2257, 6236, 5661, 11, 5884, 36781, 18032, 8244, 39781, 43076, 9319, 2289, 1543, 50915], "temperature": 0, "avg_logprob": -0.11965538340912768, "compression_ratio": 1.4297994269340975, "no_speech_prob": 1.5087918761591546e-11}, {"id": 33, "seek": 23700, "start": 248.0, "end": 251.0, "text": "第一次能射成成這樣,我覺得已經很厲害了。", "tokens": [50915, 18049, 9487, 8225, 1530, 226, 11336, 11336, 8377, 11, 17689, 25338, 4563, 20939, 2289, 1543, 51065], "temperature": 0, "avg_logprob": -0.11965538340912768, "compression_ratio": 1.4297994269340975, "no_speech_prob": 1.5087918761591546e-11}, {"id": 34, "seek": 23700, "start": 251.0, "end": 253.0, "text": "然後我們來修改一下其中的一些問題。", "tokens": [51065, 10213, 5884, 3763, 7792, 106, 34490, 8861, 9572, 5975, 1546, 38515, 17197, 1543, 51165], "temperature": 0, "avg_logprob": -0.11965538340912768, "compression_ratio": 1.4297994269340975, "no_speech_prob": 1.5087918761591546e-11}, {"id": 35, "seek": 23700, "start": 253.0, "end": 258.0, "text": "比如說這個櫻花樹的樹葉它太分散了,包括樹枝的朝向不太正確。", "tokens": [51165, 36757, 4622, 6287, 162, 104, 119, 20127, 47169, 1546, 47169, 46290, 11284, 9455, 6627, 7017, 96, 2289, 11, 41828, 47169, 7360, 251, 1546, 46610, 24282, 1960, 9455, 15789, 24293, 1543, 51415], "temperature": 0, "avg_logprob": -0.11965538340912768, "compression_ratio": 1.4297994269340975, "no_speech_prob": 1.5087918761591546e-11}, {"id": 36, "seek": 23700, "start": 258.0, "end": 261.0, "text": "完成後,我們打開渲染。", "tokens": [51415, 41509, 5661, 11, 5884, 12467, 8949, 9592, 110, 46568, 1543, 51565], "temperature": 0, "avg_logprob": -0.11965538340912768, "compression_ratio": 1.4297994269340975, "no_speech_prob": 1.5087918761591546e-11}, {"id": 37, "seek": 23700, "start": 261.0, "end": 264.0, "text": "然後可以看到它這個沒有陰影啊。", "tokens": [51565, 10213, 6723, 18032, 11284, 6287, 6963, 8842, 108, 16820, 4905, 1543, 51715], "temperature": 0, "avg_logprob": -0.11965538340912768, "compression_ratio": 1.4297994269340975, "no_speech_prob": 1.5087918761591546e-11}, {"id": 38, "seek": 23700, "start": 264.0, "end": 266.0, "text": "這個其實是它材質的問題。", "tokens": [51715, 6287, 14139, 1541, 11284, 4422, 238, 43450, 1546, 17197, 1543, 51815], "temperature": 0, "avg_logprob": -0.11965538340912768, "compression_ratio": 1.4297994269340975, "no_speech_prob": 1.5087918761591546e-11}, {"id": 39, "seek": 26600, "start": 266.0, "end": 268.0, "text": "我們讓它修改一下。", "tokens": [50365, 5884, 21195, 11284, 7792, 106, 34490, 8861, 1543, 50465], "temperature": 0, "avg_logprob": -0.10531065437231171, "compression_ratio": 1.3959537572254335, "no_speech_prob": 1.6568942398653697e-11}, {"id": 40, "seek": 26600, "start": 268.0, "end": 271.0, "text": "可以看到AI居然先給我們做了一個測試。", "tokens": [50465, 6723, 18032, 48698, 9636, 227, 5823, 10108, 17798, 5884, 10907, 2289, 8990, 9592, 105, 22099, 1543, 50615], "temperature": 0, "avg_logprob": -0.10531065437231171, "compression_ratio": 1.3959537572254335, "no_speech_prob": 1.6568942398653697e-11}, {"id": 41, "seek": 26600, "start": 271.0, "end": 273.0, "text": "真的是有夠嚴謹的。", "tokens": [50615, 45698, 2412, 31649, 29593, 112, 4802, 117, 1546, 1543, 50715], "temperature": 0, "avg_logprob": -0.10531065437231171, "compression_ratio": 1.3959537572254335, "no_speech_prob": 1.6568942398653697e-11}, {"id": 42, "seek": 26600, "start": 273.0, "end": 276.0, "text": "可以看到現在這個光影效果就有了。", "tokens": [50715, 6723, 18032, 12648, 6287, 20690, 16820, 43076, 9319, 3111, 2412, 2289, 1543, 50865], "temperature": 0, "avg_logprob": -0.10531065437231171, "compression_ratio": 1.3959537572254335, "no_speech_prob": 1.6568942398653697e-11}, {"id": 43, "seek": 26600, "start": 276.0, "end": 279.0, "text": "不過啊,三軒二的那種效果還是不夠強烈。", "tokens": [50865, 28193, 4905, 11, 10960, 14454, 240, 11217, 1546, 4184, 16516, 43076, 9319, 25583, 1960, 31649, 22752, 23661, 230, 1543, 51015], "temperature": 0, "avg_logprob": -0.10531065437231171, "compression_ratio": 1.3959537572254335, "no_speech_prob": 1.6568942398653697e-11}, {"id": 44, "seek": 26600, "start": 279.0, "end": 281.0, "text": "尤其是沒有描邊。", "tokens": [51015, 1530, 97, 9572, 1541, 6963, 11673, 237, 14901, 1543, 51115], "temperature": 0, "avg_logprob": -0.10531065437231171, "compression_ratio": 1.3959537572254335, "no_speech_prob": 1.6568942398653697e-11}, {"id": 45, "seek": 26600, "start": 281.0, "end": 283.0, "text": "這裡我們再讓它修改一下。", "tokens": [51115, 18771, 5884, 8623, 21195, 11284, 7792, 106, 34490, 8861, 1543, 51215], "temperature": 0, "avg_logprob": -0.10531065437231171, "compression_ratio": 1.3959537572254335, "no_speech_prob": 1.6568942398653697e-11}, {"id": 46, "seek": 26600, "start": 283.0, "end": 286.0, "text": "好,這下看起來顏色風格什麼的就對了嘛。", "tokens": [51215, 2131, 11, 2664, 4438, 4200, 21670, 10178, 237, 17673, 22713, 30921, 7598, 1546, 3111, 45839, 20722, 1543, 51365], "temperature": 0, "avg_logprob": -0.10531065437231171, "compression_ratio": 1.3959537572254335, "no_speech_prob": 1.6568942398653697e-11}, {"id": 47, "seek": 26600, "start": 286.0, "end": 290.0, "text": "我發現自行系的前輪胎的內胎的旋轉方向有點問題。", "tokens": [51365, 1654, 38927, 9722, 8082, 25368, 1546, 8945, 12290, 103, 19095, 236, 1546, 28472, 19095, 236, 1546, 4479, 233, 32739, 9249, 24282, 42440, 17197, 1543, 51565], "temperature": 0, "avg_logprob": -0.10531065437231171, "compression_ratio": 1.3959537572254335, "no_speech_prob": 1.6568942398653697e-11}, {"id": 48, "seek": 26600, "start": 290.0, "end": 292.0, "text": "這時候就發揮Blender的優勢了。", "tokens": [51565, 2664, 14010, 3111, 14637, 11673, 106, 14520, 3216, 1546, 12827, 103, 7978, 95, 2289, 1543, 51665], "temperature": 0, "avg_logprob": -0.10531065437231171, "compression_ratio": 1.3959537572254335, "no_speech_prob": 1.6568942398653697e-11}, {"id": 49, "seek": 26600, "start": 292.0, "end": 295.0, "text": "我們直接在Blender調整參數來修改啊。", "tokens": [51665, 5884, 43297, 3581, 14520, 3216, 32943, 27662, 47375, 30622, 3763, 7792, 106, 34490, 4905, 1543, 51815], "temperature": 0, "avg_logprob": -0.10531065437231171, "compression_ratio": 1.3959537572254335, "no_speech_prob": 1.6568942398653697e-11}, {"id": 50, "seek": 29500, "start": 295.0, "end": 297.0, "text": "我覺得便電店旁邊這個地方有點太空曠了。", "tokens": [50365, 17689, 27364, 20545, 22469, 4479, 223, 14901, 6287, 30146, 42440, 9455, 23322, 9531, 254, 2289, 1543, 50465], "temperature": 0, "avg_logprob": -0.12351053953170776, "compression_ratio": 1.0656565656565657, "no_speech_prob": 2.0113339813110365e-11}, {"id": 51, "seek": 29500, "start": 297.0, "end": 299.0, "text": "然後我讓它先登一個小汽車。", "tokens": [50465, 10213, 1654, 21195, 11284, 10108, 25874, 8990, 7322, 12800, 121, 25941, 1543, 50565], "temperature": 0, "avg_logprob": -0.12351053953170776, "compression_ratio": 1.0656565656565657, "no_speech_prob": 2.0113339813110365e-11}, {"id": 52, "seek": 29500, "start": 299.0, "end": 301.0, "text": "然後加上一個廣告牌。", "tokens": [50565, 10213, 9990, 5708, 8990, 17527, 96, 16846, 6935, 234, 1543, 50665], "temperature": 0, "avg_logprob": -0.12351053953170776, "compression_ratio": 1.0656565656565657, "no_speech_prob": 2.0113339813110365e-11}, {"id": 53, "seek": 29500, "start": 301.0, "end": 308.0, "text": "本地小模型操作Blender建模的效果和GPD6相比,你覺得有多少差距呢?", "tokens": [50665, 8802, 10928, 7322, 41908, 39823, 13167, 235, 11914, 14520, 3216, 34157, 41908, 1546, 43076, 9319, 12565, 38, 47, 35, 21, 15106, 11706, 11, 2166, 11274, 2412, 41492, 21679, 6563, 251, 6240, 30, 51015], "temperature": 0, "avg_logprob": -0.12351053953170776, "compression_ratio": 1.0656565656565657, "no_speech_prob": 2.0113339813110365e-11}, {"id": 54, "seek": 30800, "start": 308.0, "end": 313.0, "text": "想用千聞3.8的漆皮建模,其專業領域還是在3.js上。", "tokens": [50365, 7093, 9254, 20787, 25468, 18, 13, 23, 1546, 14065, 228, 34582, 34157, 41908, 11, 9572, 46072, 27119, 8313, 246, 16262, 253, 25583, 3581, 18, 13, 25530, 5708, 1543, 50615], "temperature": 0, "avg_logprob": -0.13828171000761144, "compression_ratio": 1.1291666666666667, "no_speech_prob": 1.9139249213262488e-11}, {"id": 55, "seek": 30800, "start": 313.0, "end": 318.0, "text": "大家看,這是我用同樣的提示詞,只不過使用的是3.js生成的。", "tokens": [50615, 6868, 4200, 11, 16600, 1654, 9254, 13089, 6180, 1546, 20949, 25696, 6716, 252, 11, 14003, 28193, 22982, 9254, 24620, 18, 13, 25530, 8244, 11336, 1546, 1543, 50865], "temperature": 0, "avg_logprob": -0.13828171000761144, "compression_ratio": 1.1291666666666667, "no_speech_prob": 1.9139249213262488e-11}, {"id": 56, "seek": 30800, "start": 318.0, "end": 320.0, "text": "是不是效果提升一大截呢?", "tokens": [50865, 23034, 43076, 9319, 20949, 41670, 2257, 3582, 1486, 103, 6240, 30, 50965], "temperature": 0, "avg_logprob": -0.13828171000761144, "compression_ratio": 1.1291666666666667, "no_speech_prob": 1.9139249213262488e-11}, {"id": 57, "seek": 30800, "start": 320.0, "end": 326.0, "text": "實際上,如果你會用Blender,你可以簡單調整,就可以獲得更好的渲染效果。", "tokens": [50965, 10376, 34837, 5708, 11, 45669, 6236, 9254, 14520, 3216, 11, 42766, 31995, 32943, 27662, 11, 36781, 29057, 110, 5916, 19002, 20715, 9592, 110, 46568, 43076, 9319, 1543, 51265], "temperature": 0, "avg_logprob": -0.13828171000761144, "compression_ratio": 1.1291666666666667, "no_speech_prob": 1.9139249213262488e-11}, {"id": 58, "seek": 32600, "start": 326.0, "end": 329.0, "text": "也就是說,Blender其實上限更高。", "tokens": [50365, 6404, 26222, 11, 14520, 3216, 14139, 5708, 43446, 19002, 12979, 1543, 50515], "temperature": 0, "avg_logprob": -0.0984995790890285, "compression_ratio": 0.9615384615384616, "no_speech_prob": 2.060454237784448e-11}, {"id": 59, "seek": 32600, "start": 329.0, "end": 333.0, "text": "但是如果你只是在網頁上實現一些程序化的生成的模型。", "tokens": [50515, 11189, 45669, 36859, 3581, 26352, 8313, 223, 5708, 10376, 9581, 38515, 29649, 6346, 237, 23756, 1546, 8244, 11336, 1546, 41908, 39823, 1543, 50715], "temperature": 0, "avg_logprob": -0.0984995790890285, "compression_ratio": 0.9615384615384616, "no_speech_prob": 2.060454237784448e-11}, {"id": 60, "seek": 32600, "start": 333.0, "end": 336.0, "text": "那麼3.js仍然是優先選擇。", "tokens": [50715, 20504, 18, 13, 25530, 1550, 235, 5823, 1541, 12827, 103, 10108, 21625, 13167, 229, 1543, 50865], "temperature": 0, "avg_logprob": -0.0984995790890285, "compression_ratio": 0.9615384615384616, "no_speech_prob": 2.060454237784448e-11}, {"id": 61, "seek": 33600, "start": 336.0, "end": 342.0, "text": "因為直接生成3.js,比起用AI來操作Blender的速度要快速唄。", "tokens": [50365, 11471, 43297, 8244, 11336, 18, 13, 25530, 11, 11706, 9147, 9254, 48698, 3763, 13167, 235, 11914, 14520, 3216, 1546, 31217, 13127, 4275, 10251, 31217, 17198, 226, 1543, 50665], "temperature": 0, "avg_logprob": -0.12402962301379052, "compression_ratio": 1.1541501976284585, "no_speech_prob": 1.6639621971958896e-11}, {"id": 62, "seek": 33600, "start": 342.0, "end": 348.0, "text": "如果你想了解如何使用3.js建模,為你的網站搭接一些3D元素或動向。", "tokens": [50665, 13119, 38386, 2289, 17278, 43526, 22982, 9254, 18, 13, 25530, 34157, 41908, 11, 6344, 18961, 26352, 34155, 22220, 255, 14468, 38515, 18, 35, 14812, 37204, 19780, 12572, 24282, 1543, 50965], "temperature": 0, "avg_logprob": -0.12402962301379052, "compression_ratio": 1.1541501976284585, "no_speech_prob": 1.6639621971958896e-11}, {"id": 63, "seek": 33600, "start": 348.0, "end": 350.0, "text": "關注我吧。", "tokens": [50965, 14899, 26432, 1654, 6062, 1543, 51065], "temperature": 0, "avg_logprob": -0.12402962301379052, "compression_ratio": 1.1541501976284585, "no_speech_prob": 1.6639621971958896e-11}, {"id": 64, "seek": 33600, "start": 350.0, "end": 353.0, "text": "後續我將分享如何使用3.js制作這樣的場景。", "tokens": [51065, 5661, 29583, 1654, 34168, 30855, 43526, 22982, 9254, 18, 13, 25530, 25491, 11914, 39425, 21446, 50218, 1543, 51215], "temperature": 0, "avg_logprob": -0.12402962301379052, "compression_ratio": 1.1541501976284585, "no_speech_prob": 1.6639621971958896e-11}, {"id": 65, "seek": 33600, "start": 353.0, "end": 355.0, "text": "感謝大家看到這裡。", "tokens": [51215, 40143, 6868, 18032, 18771, 1543, 51315], "temperature": 0, "avg_logprob": -0.12402962301379052, "compression_ratio": 1.1541501976284585, "no_speech_prob": 1.6639621971958896e-11}, {"id": 66, "seek": 33600, "start": 355.0, "end": 357.0, "text": "我是ZZ,期待下次相遇。", "tokens": [51315, 15914, 57, 57, 11, 16786, 18390, 35829, 15106, 3330, 229, 1543, 51415], "temperature": 0, "avg_logprob": -0.12402962301379052, "compression_ratio": 1.1541501976284585, "no_speech_prob": 1.6639621971958896e-11}], "language": "zh"}