跳转至内容
  • 版块
  • 最新
  • 标签
  • 热门
  • 用户
  • 群组
皮肤
  • 浅色
  • Brite
  • Cerulean
  • Cosmo
  • Flatly
  • Journal
  • Litera
  • Lumen
  • Lux
  • Materia
  • Minty
  • Morph
  • Pulse
  • Sandstone
  • Simplex
  • Sketchy
  • Spacelab
  • United
  • Yeti
  • Zephyr
  • 深色
  • Cyborg
  • Darkly
  • Quartz
  • Slate
  • Solar
  • Superhero
  • Vapor

  • 默认(LCZ-Blue)
  • 不使用皮肤
  • LCZ-Green
  • LCZ-Blue
  • LCZ-Black
折叠
品牌标识

抡锤者

首页 版块 标签
硬件
装备秀
AI 广场 App
  1. 主页
  2. 版块
  3. AI音视频画图
  4. 转场片段之一,H3

转场片段之一,H3

已定时 已固定 已锁定 已移动 AI音视频画图
视频生成minimax
13 帖子 3 发布者 235 浏览
  • 从旧到新
  • 从新到旧
  • 最多赞同
回复
  • 在新帖中回复
登录后回复
此主题已被删除。只有拥有主题管理权限的用户可以查看。
  • Tony WangT Tony Wang

    @Thanaots

    我不太懂视频. 不过我看 005 和 010 不是挺不错的吗? 尾帧是不是本来就尺寸和缩放比例不对?

    T
    T
    Thanaots
    体面乡绅 硕果累累
    编写于 最后由 编辑
    #3

    @Tony-Wang 说:

    @Thanaots

    我不太懂视频. 不过我看 005 和 010 不是挺不错的吗? 尾帧是不是本来就尺寸和缩放比例不对?

    运镜还可以。但是问题是我要求的内容,没有办法完全展现出来。最大问题是i2v,对我的语音理解不够细致。

    1 条回复 最后回复
    0
    • T Thanaots

      Youtube Video

      我的一个转场片段,整体来说做的不是很成功。我现在正在尝试用win2.2重置,然后底下是我做这个片段时候的一些个详细信息
      生成记录.txt

      imbiplaza ASUSI
      imbiplaza ASUSI
      imbiplaza ASUS
      谪仙人
      编写于 最后由 编辑
      #4

      @Thanaots

      我是来乱的。。。。

      Youtube Video

      提示词:
      Masterpiece, movie transition video, continuous endless zoom-in camera movement. Starting with a macro shot zooming into a person's black sunglasses reflection, seamlessly transitioning into a cyberpunk megacity with neon holographic billboards and flying cars. Camera keeps moving forward, zooming into a second person's retro black sunglasses on the cyberpunk street, flawlessly piercing through the lens to reveal a nostalgic 1980s Hong Kong street with warm moody lighting and double-decker buses. The camera never stops, deeply zooming into a driver's aviator sunglasses on the Hong Kong street, seamlessly transitioning into a wide shot of an epic desert highway road trip down Route 66. Camera maintains steady forward momentum, zooming into a pedestrian's round black sunglasses by the road, the lens reflection warps like ripples into swirling liquid colors, opening up into a surreal psychedelic dreamscape with melting clocks. Finally, the camera thrusts forward into sci-fi black goggles within the dream, flashing bright green UI scanning lines, seamlessly diving deep into an endless 3D digital matrix space with flowing cyber grids and energetic light streaks. Flawless transitions throughout, hyper-realistic, 8k resolution, cinematic motion blur, immersive visual journey.

      整合提示詞的設計邏輯與轉場順序:

      為了讓 AI 理解這是一段連續的長鏡頭,提示詞內部建立了一條「墨鏡 ➔ 場景 ➔ 場景中新人物的墨鏡 ➔ 下一個場景」的循環遞進邏輯:

      1. 起點:從初始人物的黑色墨鏡反射出發。
      2. 第一段穿梭:穿過墨鏡 ➔ 進入【賽博朋克都市】。
      3. 第二段穿梭:鏡頭在賽博都市中推進,鎖定新人物的粗框墨鏡 ➔ 穿過鏡片進入【80年代復古香港街頭】。
      4. 第三段穿梭:鏡頭在香港街頭繼續前推,鎖定司機的飛行員墨鏡 ➔ 穿過鏡片進入【美國66號荒漠公路】。
      5. 第四段穿梭:鏡頭在公路旁前推,鎖定路人的圓框墨鏡,鏡面化為彩色水波 ➔ 穿過水波進入【達利風迷幻幾何空間】。
      6. 第五段穿梭:鏡頭在迷幻空間中最後衝刺,衝入未來科技防護鏡 ➔ 最終進入【3D虛擬數位矩陣空間】結尾。
      T 1 条回复 最后回复
      2
      • imbiplaza ASUSI imbiplaza ASUS

        @Thanaots

        我是来乱的。。。。

        Youtube Video

        提示词:
        Masterpiece, movie transition video, continuous endless zoom-in camera movement. Starting with a macro shot zooming into a person's black sunglasses reflection, seamlessly transitioning into a cyberpunk megacity with neon holographic billboards and flying cars. Camera keeps moving forward, zooming into a second person's retro black sunglasses on the cyberpunk street, flawlessly piercing through the lens to reveal a nostalgic 1980s Hong Kong street with warm moody lighting and double-decker buses. The camera never stops, deeply zooming into a driver's aviator sunglasses on the Hong Kong street, seamlessly transitioning into a wide shot of an epic desert highway road trip down Route 66. Camera maintains steady forward momentum, zooming into a pedestrian's round black sunglasses by the road, the lens reflection warps like ripples into swirling liquid colors, opening up into a surreal psychedelic dreamscape with melting clocks. Finally, the camera thrusts forward into sci-fi black goggles within the dream, flashing bright green UI scanning lines, seamlessly diving deep into an endless 3D digital matrix space with flowing cyber grids and energetic light streaks. Flawless transitions throughout, hyper-realistic, 8k resolution, cinematic motion blur, immersive visual journey.

        整合提示詞的設計邏輯與轉場順序:

        為了讓 AI 理解這是一段連續的長鏡頭,提示詞內部建立了一條「墨鏡 ➔ 場景 ➔ 場景中新人物的墨鏡 ➔ 下一個場景」的循環遞進邏輯:

        1. 起點:從初始人物的黑色墨鏡反射出發。
        2. 第一段穿梭:穿過墨鏡 ➔ 進入【賽博朋克都市】。
        3. 第二段穿梭:鏡頭在賽博都市中推進,鎖定新人物的粗框墨鏡 ➔ 穿過鏡片進入【80年代復古香港街頭】。
        4. 第三段穿梭:鏡頭在香港街頭繼續前推,鎖定司機的飛行員墨鏡 ➔ 穿過鏡片進入【美國66號荒漠公路】。
        5. 第四段穿梭:鏡頭在公路旁前推,鎖定路人的圓框墨鏡,鏡面化為彩色水波 ➔ 穿過水波進入【達利風迷幻幾何空間】。
        6. 第五段穿梭:鏡頭在迷幻空間中最後衝刺,衝入未來科技防護鏡 ➔ 最終進入【3D虛擬數位矩陣空間】結尾。
        T
        T
        Thanaots
        体面乡绅 硕果累累
        编写于 最后由 编辑
        #5

        @imbiplaza-ASUS 说:

        @Thanaots

        我是来乱的。。。。

        Youtube Video

        提示词:
        Masterpiece, movie transition video, continuous endless zoom-in camera movement. Starting with a macro shot zooming into a person's black sunglasses reflection, seamlessly transitioning into a cyberpunk megacity with neon holographic billboards and flying cars. Camera keeps moving forward, zooming into a second person's retro black sunglasses on the cyberpunk street, flawlessly piercing through the lens to reveal a nostalgic 1980s Hong Kong street with warm moody lighting and double-decker buses. The camera never stops, deeply zooming into a driver's aviator sunglasses on the Hong Kong street, seamlessly transitioning into a wide shot of an epic desert highway road trip down Route 66. Camera maintains steady forward momentum, zooming into a pedestrian's round black sunglasses by the road, the lens reflection warps like ripples into swirling liquid colors, opening up into a surreal psychedelic dreamscape with melting clocks. Finally, the camera thrusts forward into sci-fi black goggles within the dream, flashing bright green UI scanning lines, seamlessly diving deep into an endless 3D digital matrix space with flowing cyber grids and energetic light streaks. Flawless transitions throughout, hyper-realistic, 8k resolution, cinematic motion blur, immersive visual journey.

        整合提示詞的設計邏輯與轉場順序:

        為了讓 AI 理解這是一段連續的長鏡頭,提示詞內部建立了一條「墨鏡 ➔ 場景 ➔ 場景中新人物的墨鏡 ➔ 下一個場景」的循環遞進邏輯:

        1. 起點:從初始人物的黑色墨鏡反射出發。
        2. 第一段穿梭:穿過墨鏡 ➔ 進入【賽博朋克都市】。
        3. 第二段穿梭:鏡頭在賽博都市中推進,鎖定新人物的粗框墨鏡 ➔ 穿過鏡片進入【80年代復古香港街頭】。
        4. 第三段穿梭:鏡頭在香港街頭繼續前推,鎖定司機的飛行員墨鏡 ➔ 穿過鏡片進入【美國66號荒漠公路】。
        5. 第四段穿梭:鏡頭在公路旁前推,鎖定路人的圓框墨鏡,鏡面化為彩色水波 ➔ 穿過水波進入【達利風迷幻幾何空間】。
        6. 第五段穿梭:鏡頭在迷幻空間中最後衝刺,衝入未來科技防護鏡 ➔ 最終進入【3D虛擬數位矩陣空間】結尾。

        大佬可以呀。
        主要是我给的提示词已经很不错了。一开始也有能勉强看的视频,比如05,09这种。但是现实逻辑不对呀。哪有长椅在马路中间的😂
        我现在不知道是seed出问题了。还是提示词出问题。
        Hermes傻逼了,现在总是给我侧身画面,我说了一万遍,都是给我侧身。。。。。我人麻了。

        imbiplaza ASUSI 1 条回复 最后回复
        1
        • T Thanaots

          @imbiplaza-ASUS 说:

          @Thanaots

          我是来乱的。。。。

          Youtube Video

          提示词:
          Masterpiece, movie transition video, continuous endless zoom-in camera movement. Starting with a macro shot zooming into a person's black sunglasses reflection, seamlessly transitioning into a cyberpunk megacity with neon holographic billboards and flying cars. Camera keeps moving forward, zooming into a second person's retro black sunglasses on the cyberpunk street, flawlessly piercing through the lens to reveal a nostalgic 1980s Hong Kong street with warm moody lighting and double-decker buses. The camera never stops, deeply zooming into a driver's aviator sunglasses on the Hong Kong street, seamlessly transitioning into a wide shot of an epic desert highway road trip down Route 66. Camera maintains steady forward momentum, zooming into a pedestrian's round black sunglasses by the road, the lens reflection warps like ripples into swirling liquid colors, opening up into a surreal psychedelic dreamscape with melting clocks. Finally, the camera thrusts forward into sci-fi black goggles within the dream, flashing bright green UI scanning lines, seamlessly diving deep into an endless 3D digital matrix space with flowing cyber grids and energetic light streaks. Flawless transitions throughout, hyper-realistic, 8k resolution, cinematic motion blur, immersive visual journey.

          整合提示詞的設計邏輯與轉場順序:

          為了讓 AI 理解這是一段連續的長鏡頭,提示詞內部建立了一條「墨鏡 ➔ 場景 ➔ 場景中新人物的墨鏡 ➔ 下一個場景」的循環遞進邏輯:

          1. 起點:從初始人物的黑色墨鏡反射出發。
          2. 第一段穿梭:穿過墨鏡 ➔ 進入【賽博朋克都市】。
          3. 第二段穿梭:鏡頭在賽博都市中推進,鎖定新人物的粗框墨鏡 ➔ 穿過鏡片進入【80年代復古香港街頭】。
          4. 第三段穿梭:鏡頭在香港街頭繼續前推,鎖定司機的飛行員墨鏡 ➔ 穿過鏡片進入【美國66號荒漠公路】。
          5. 第四段穿梭:鏡頭在公路旁前推,鎖定路人的圓框墨鏡,鏡面化為彩色水波 ➔ 穿過水波進入【達利風迷幻幾何空間】。
          6. 第五段穿梭:鏡頭在迷幻空間中最後衝刺,衝入未來科技防護鏡 ➔ 最終進入【3D虛擬數位矩陣空間】結尾。

          大佬可以呀。
          主要是我给的提示词已经很不错了。一开始也有能勉强看的视频,比如05,09这种。但是现实逻辑不对呀。哪有长椅在马路中间的😂
          我现在不知道是seed出问题了。还是提示词出问题。
          Hermes傻逼了,现在总是给我侧身画面,我说了一万遍,都是给我侧身。。。。。我人麻了。

          imbiplaza ASUSI
          imbiplaza ASUSI
          imbiplaza ASUS
          谪仙人
          编写于 最后由 编辑
          #6

          @Thanaots

          看看能够调出hermes 给h3 的提示词吗?我看看帮你改

          T 1 条回复 最后回复
          1
          • imbiplaza ASUSI imbiplaza ASUS

            @Thanaots

            看看能够调出hermes 给h3 的提示词吗?我看看帮你改

            T
            T
            Thanaots
            体面乡绅 硕果累累
            编写于 最后由 编辑
            #7

            @imbiplaza-ASUS 说:

            @Thanaots

            看看能够调出hermes 给h3 的提示词吗?我看看帮你改

            这是当前最新的提示词(航拍下摇 → 落地 → 推进 → 转脸看镜头),中英文:

            英文:

            A continuous unbroken single shot: the camera starts as an aerial top-down view high above a bright sunny Manhattan, then slowly and gently descends in one unhurried crane motion toward the roadside of Fifth Avenue, settling directly in front of a woman with black shoulder-length hair and black sunglasses wearing a red long-sleeve top sitting on a wooden bench placed on the sidewalk right by the curb, clearly at the roadside, absolutely not in the middle of the road, seated at one end of the bench. She faces the camera directly, body completely frontal facing the camera straight on, head fully frontal, absolutely no side profile, no three-quarter view, do not turn the body or head to either side. Her head is tilted slightly away at first, looking to the side, then as the camera very slowly pushes in closer, at the very end she slowly turns her head back to stare straight into the lens, ending the shot looking directly at the camera. Stern serious unsmiling expression, no smile. Busy sidewalk with pedestrians walking past, background slightly blurred. Golden warm sunlight, photorealistic, cinematic film look, shallow depth of field, one single continuous unbroken camera movement, absolutely no cuts, no jump cuts, no scene change, the camera glides in one slow unbroken take from start to finish, no map grid, no text, no watermark

            中文:

            一个连续不断的一镜到底镜头:镜头从明亮的曼哈顿上空的高空俯视开始,然后以舒缓、从容的摇臂运动缓缓下降,朝第五大道路边落下,最终正对一位坐在木质长椅上的女子——她黑色及肩发、戴黑色墨镜、穿红色长袖上衣;长椅摆放在路缘石旁的人行道上,明确在路边,绝不在马路中间,她坐在长椅的一端。她正面朝向镜头,身体完全正对镜头,头部完全正对,绝无侧面、绝无四分之三角度,身体或头不转向任何一侧。起初她的头微微偏向一侧、看向旁边,随着镜头非常缓慢地推进靠近,在最后她缓缓把头转回、直直凝视镜头,以直视镜头收尾。表情严肃、不苟言笑、不笑。繁忙的人行道上行人往来,背景略微虚化。金色温暖阳光,照片级真实感,电影质感,浅景深,一个单一连续不断的运镜,绝无剪辑、绝无跳切、绝无场景切换,镜头从头到尾缓慢连贯一气呵成,无地图网格、无文字、无水印。

            imbiplaza ASUSI 1 条回复 最后回复
            0
            • T Thanaots

              @imbiplaza-ASUS 说:

              @Thanaots

              看看能够调出hermes 给h3 的提示词吗?我看看帮你改

              这是当前最新的提示词(航拍下摇 → 落地 → 推进 → 转脸看镜头),中英文:

              英文:

              A continuous unbroken single shot: the camera starts as an aerial top-down view high above a bright sunny Manhattan, then slowly and gently descends in one unhurried crane motion toward the roadside of Fifth Avenue, settling directly in front of a woman with black shoulder-length hair and black sunglasses wearing a red long-sleeve top sitting on a wooden bench placed on the sidewalk right by the curb, clearly at the roadside, absolutely not in the middle of the road, seated at one end of the bench. She faces the camera directly, body completely frontal facing the camera straight on, head fully frontal, absolutely no side profile, no three-quarter view, do not turn the body or head to either side. Her head is tilted slightly away at first, looking to the side, then as the camera very slowly pushes in closer, at the very end she slowly turns her head back to stare straight into the lens, ending the shot looking directly at the camera. Stern serious unsmiling expression, no smile. Busy sidewalk with pedestrians walking past, background slightly blurred. Golden warm sunlight, photorealistic, cinematic film look, shallow depth of field, one single continuous unbroken camera movement, absolutely no cuts, no jump cuts, no scene change, the camera glides in one slow unbroken take from start to finish, no map grid, no text, no watermark

              中文:

              一个连续不断的一镜到底镜头:镜头从明亮的曼哈顿上空的高空俯视开始,然后以舒缓、从容的摇臂运动缓缓下降,朝第五大道路边落下,最终正对一位坐在木质长椅上的女子——她黑色及肩发、戴黑色墨镜、穿红色长袖上衣;长椅摆放在路缘石旁的人行道上,明确在路边,绝不在马路中间,她坐在长椅的一端。她正面朝向镜头,身体完全正对镜头,头部完全正对,绝无侧面、绝无四分之三角度,身体或头不转向任何一侧。起初她的头微微偏向一侧、看向旁边,随着镜头非常缓慢地推进靠近,在最后她缓缓把头转回、直直凝视镜头,以直视镜头收尾。表情严肃、不苟言笑、不笑。繁忙的人行道上行人往来,背景略微虚化。金色温暖阳光,照片级真实感,电影质感,浅景深,一个单一连续不断的运镜,绝无剪辑、绝无跳切、绝无场景切换,镜头从头到尾缓慢连贯一气呵成,无地图网格、无文字、无水印。

              imbiplaza ASUSI
              imbiplaza ASUSI
              imbiplaza ASUS
              谪仙人
              编写于 最后由 编辑
              #8

              @Thanaots

              generated_preview (1).gif

              這個問題是 AI 影片生成模型常出現的**「語意混淆」與「過度強調負面詞」**所導致的。
              當你在提示詞中寫了 absolutely not in the middle of the road(絕對不在馬路中間)時,AI 的注意力機制反而會捕捉到 middle of the road 這個高頻詞,進而把椅子放到了馬路中央。另外,原本的提示詞結構較長,馬路、長椅和行人的空間關係不夠明確。
              為了讓 MiniMax H3 能精準理解「長椅在人行道上,馬路在背景/旁邊」,我幫你重新優化了這段提示詞。我們將空間佈局改為正向肯定句,並強化了空間的層次感(人行道 ➔ 路緣 ➔ 車道)。

              🛠️ 優化後的 MiniMax H3 專用提示詞

              A continuous unbroken single-take crane shot. The camera begins as a high aerial top-down view above sunny Manhattan, then smoothly and slowly descends toward the wide pedestrian sidewalk of Fifth Avenue. The camera settles right on the sidewalk, directly facing a woman with black shoulder-length hair and black sunglasses, wearing a red long-sleeve top. She is seated at one end of a wooden bench that is positioned securely on the wide concrete sidewalk next to a building wall. The bustling sidewalk is filled with pedestrians walking past, while the busy driving street with cars is strictly in the background behind the sidewalk. The woman faces the camera perfectly frontal, body and head fully straight on, no side profile, no angled view. Her head is tilted slightly away looking toward the side at first, then as the camera very slowly pushes into a closer shot, she slowly turns her head back to stare directly into the camera lens with a stern, serious, unsmiling expression. Golden warm sunlight, photorealistic cinematic film look, shallow depth of field, one single seamless continuous camera movement, no cuts, no scene change, no text, no watermark.

              🔑 關鍵優化解析(為什麼這樣改?):

              1. 剔除負面詞,改用「絕對定位」:刪除了 absolutely not in the middle of the road,改用 positioned securely on the wide concrete sidewalk next to a building wall(牢固地安置在靠近建築物牆面的寬闊水泥人行道上)。這樣能給 AI 一個極其明確的物理參照物(牆面),強迫長椅遠離馬路。
              2. 明確分開「人行道」與「車道」的空間層次:使用 The bustling sidewalk is filled with pedestrians... while the busy driving street with cars is strictly in the background。明確告訴 AI,馬路和車輛只能作為「背景」存在於人行道後方。
              3. 優化鏡頭動態的連貫性描述:將開頭改為 A continuous unbroken single-take crane shot,並在中間補強 settles right on the sidewalk,引導 MiniMax H3 的鏡頭從空中降落時,直接把「落脚點」鎖定在人行道上,而不是降落到馬路中央。
                你可以直接將這段優化後的提示詞投入 MiniMax H3 測試。如果生成後發現背景的馬路車流依然太過干擾,我們可以嘗試把場景換到沒有車道交匯的純人行步道或公園邊緣,你需要我朝這個方向幫你做備用調整嗎?
              T 1 条回复 最后回复
              1
              • T
                T
                Thanaots
                体面乡绅 硕果累累
                编写于 最后由 编辑
                #9

                最后抽奖抽出来保底了,现在这一段可算是可用了

                1 条回复 最后回复
                0
                • T
                  T
                  Thanaots
                  体面乡绅 硕果累累
                  编写于 最后由 编辑
                  #10

                  14130_under_2MB.mp4

                  1 条回复 最后回复
                  0
                  • imbiplaza ASUSI imbiplaza ASUS

                    @Thanaots

                    generated_preview (1).gif

                    這個問題是 AI 影片生成模型常出現的**「語意混淆」與「過度強調負面詞」**所導致的。
                    當你在提示詞中寫了 absolutely not in the middle of the road(絕對不在馬路中間)時,AI 的注意力機制反而會捕捉到 middle of the road 這個高頻詞,進而把椅子放到了馬路中央。另外,原本的提示詞結構較長,馬路、長椅和行人的空間關係不夠明確。
                    為了讓 MiniMax H3 能精準理解「長椅在人行道上,馬路在背景/旁邊」,我幫你重新優化了這段提示詞。我們將空間佈局改為正向肯定句,並強化了空間的層次感(人行道 ➔ 路緣 ➔ 車道)。

                    🛠️ 優化後的 MiniMax H3 專用提示詞

                    A continuous unbroken single-take crane shot. The camera begins as a high aerial top-down view above sunny Manhattan, then smoothly and slowly descends toward the wide pedestrian sidewalk of Fifth Avenue. The camera settles right on the sidewalk, directly facing a woman with black shoulder-length hair and black sunglasses, wearing a red long-sleeve top. She is seated at one end of a wooden bench that is positioned securely on the wide concrete sidewalk next to a building wall. The bustling sidewalk is filled with pedestrians walking past, while the busy driving street with cars is strictly in the background behind the sidewalk. The woman faces the camera perfectly frontal, body and head fully straight on, no side profile, no angled view. Her head is tilted slightly away looking toward the side at first, then as the camera very slowly pushes into a closer shot, she slowly turns her head back to stare directly into the camera lens with a stern, serious, unsmiling expression. Golden warm sunlight, photorealistic cinematic film look, shallow depth of field, one single seamless continuous camera movement, no cuts, no scene change, no text, no watermark.

                    🔑 關鍵優化解析(為什麼這樣改?):

                    1. 剔除負面詞,改用「絕對定位」:刪除了 absolutely not in the middle of the road,改用 positioned securely on the wide concrete sidewalk next to a building wall(牢固地安置在靠近建築物牆面的寬闊水泥人行道上)。這樣能給 AI 一個極其明確的物理參照物(牆面),強迫長椅遠離馬路。
                    2. 明確分開「人行道」與「車道」的空間層次:使用 The bustling sidewalk is filled with pedestrians... while the busy driving street with cars is strictly in the background。明確告訴 AI,馬路和車輛只能作為「背景」存在於人行道後方。
                    3. 優化鏡頭動態的連貫性描述:將開頭改為 A continuous unbroken single-take crane shot,並在中間補強 settles right on the sidewalk,引導 MiniMax H3 的鏡頭從空中降落時,直接把「落脚點」鎖定在人行道上,而不是降落到馬路中央。
                      你可以直接將這段優化後的提示詞投入 MiniMax H3 測試。如果生成後發現背景的馬路車流依然太過干擾,我們可以嘗試把場景換到沒有車道交匯的純人行步道或公園邊緣,你需要我朝這個方向幫你做備用調整嗎?
                    T
                    T
                    Thanaots
                    体面乡绅 硕果累累
                    编写于 最后由 编辑
                    #11

                    @imbiplaza-ASUS 说:

                    @Thanaots

                    generated_preview (1).gif

                    這個問題是 AI 影片生成模型常出現的**「語意混淆」與「過度強調負面詞」**所導致的。
                    當你在提示詞中寫了 absolutely not in the middle of the road(絕對不在馬路中間)時,AI 的注意力機制反而會捕捉到 middle of the road 這個高頻詞,進而把椅子放到了馬路中央。另外,原本的提示詞結構較長,馬路、長椅和行人的空間關係不夠明確。
                    為了讓 MiniMax H3 能精準理解「長椅在人行道上,馬路在背景/旁邊」,我幫你重新優化了這段提示詞。我們將空間佈局改為正向肯定句,並強化了空間的層次感(人行道 ➔ 路緣 ➔ 車道)。

                    🛠️ 優化後的 MiniMax H3 專用提示詞

                    A continuous unbroken single-take crane shot. The camera begins as a high aerial top-down view above sunny Manhattan, then smoothly and slowly descends toward the wide pedestrian sidewalk of Fifth Avenue. The camera settles right on the sidewalk, directly facing a woman with black shoulder-length hair and black sunglasses, wearing a red long-sleeve top. She is seated at one end of a wooden bench that is positioned securely on the wide concrete sidewalk next to a building wall. The bustling sidewalk is filled with pedestrians walking past, while the busy driving street with cars is strictly in the background behind the sidewalk. The woman faces the camera perfectly frontal, body and head fully straight on, no side profile, no angled view. Her head is tilted slightly away looking toward the side at first, then as the camera very slowly pushes into a closer shot, she slowly turns her head back to stare directly into the camera lens with a stern, serious, unsmiling expression. Golden warm sunlight, photorealistic cinematic film look, shallow depth of field, one single seamless continuous camera movement, no cuts, no scene change, no text, no watermark.

                    🔑 關鍵優化解析(為什麼這樣改?):

                    1. 剔除負面詞,改用「絕對定位」:刪除了 absolutely not in the middle of the road,改用 positioned securely on the wide concrete sidewalk next to a building wall(牢固地安置在靠近建築物牆面的寬闊水泥人行道上)。這樣能給 AI 一個極其明確的物理參照物(牆面),強迫長椅遠離馬路。
                    2. 明確分開「人行道」與「車道」的空間層次:使用 The bustling sidewalk is filled with pedestrians... while the busy driving street with cars is strictly in the background。明確告訴 AI,馬路和車輛只能作為「背景」存在於人行道後方。
                    3. 優化鏡頭動態的連貫性描述:將開頭改為 A continuous unbroken single-take crane shot,並在中間補強 settles right on the sidewalk,引導 MiniMax H3 的鏡頭從空中降落時,直接把「落脚點」鎖定在人行道上,而不是降落到馬路中央。
                      你可以直接將這段優化後的提示詞投入 MiniMax H3 測試。如果生成後發現背景的馬路車流依然太過干擾,我們可以嘗試把場景換到沒有車道交匯的純人行步道或公園邊緣,你需要我朝這個方向幫你做備用調整嗎?

                    大佬,怎么感觉提示词也不过是抽奖的一部分?

                    imbiplaza ASUSI 1 条回复 最后回复
                    0
                    • T Thanaots

                      @imbiplaza-ASUS 说:

                      @Thanaots

                      generated_preview (1).gif

                      這個問題是 AI 影片生成模型常出現的**「語意混淆」與「過度強調負面詞」**所導致的。
                      當你在提示詞中寫了 absolutely not in the middle of the road(絕對不在馬路中間)時,AI 的注意力機制反而會捕捉到 middle of the road 這個高頻詞,進而把椅子放到了馬路中央。另外,原本的提示詞結構較長,馬路、長椅和行人的空間關係不夠明確。
                      為了讓 MiniMax H3 能精準理解「長椅在人行道上,馬路在背景/旁邊」,我幫你重新優化了這段提示詞。我們將空間佈局改為正向肯定句,並強化了空間的層次感(人行道 ➔ 路緣 ➔ 車道)。

                      🛠️ 優化後的 MiniMax H3 專用提示詞

                      A continuous unbroken single-take crane shot. The camera begins as a high aerial top-down view above sunny Manhattan, then smoothly and slowly descends toward the wide pedestrian sidewalk of Fifth Avenue. The camera settles right on the sidewalk, directly facing a woman with black shoulder-length hair and black sunglasses, wearing a red long-sleeve top. She is seated at one end of a wooden bench that is positioned securely on the wide concrete sidewalk next to a building wall. The bustling sidewalk is filled with pedestrians walking past, while the busy driving street with cars is strictly in the background behind the sidewalk. The woman faces the camera perfectly frontal, body and head fully straight on, no side profile, no angled view. Her head is tilted slightly away looking toward the side at first, then as the camera very slowly pushes into a closer shot, she slowly turns her head back to stare directly into the camera lens with a stern, serious, unsmiling expression. Golden warm sunlight, photorealistic cinematic film look, shallow depth of field, one single seamless continuous camera movement, no cuts, no scene change, no text, no watermark.

                      🔑 關鍵優化解析(為什麼這樣改?):

                      1. 剔除負面詞,改用「絕對定位」:刪除了 absolutely not in the middle of the road,改用 positioned securely on the wide concrete sidewalk next to a building wall(牢固地安置在靠近建築物牆面的寬闊水泥人行道上)。這樣能給 AI 一個極其明確的物理參照物(牆面),強迫長椅遠離馬路。
                      2. 明確分開「人行道」與「車道」的空間層次:使用 The bustling sidewalk is filled with pedestrians... while the busy driving street with cars is strictly in the background。明確告訴 AI,馬路和車輛只能作為「背景」存在於人行道後方。
                      3. 優化鏡頭動態的連貫性描述:將開頭改為 A continuous unbroken single-take crane shot,並在中間補強 settles right on the sidewalk,引導 MiniMax H3 的鏡頭從空中降落時,直接把「落脚點」鎖定在人行道上,而不是降落到馬路中央。
                        你可以直接將這段優化後的提示詞投入 MiniMax H3 測試。如果生成後發現背景的馬路車流依然太過干擾,我們可以嘗試把場景換到沒有車道交匯的純人行步道或公園邊緣,你需要我朝這個方向幫你做備用調整嗎?

                      大佬,怎么感觉提示词也不过是抽奖的一部分?

                      imbiplaza ASUSI
                      imbiplaza ASUSI
                      imbiplaza ASUS
                      谪仙人
                      编写于 最后由 编辑
                      #12

                      @Thanaots

                      提示词抽奖 + 视频生成抽奖

                      T 1 条回复 最后回复
                      0
                      • imbiplaza ASUSI imbiplaza ASUS

                        @Thanaots

                        提示词抽奖 + 视频生成抽奖

                        T
                        T
                        Thanaots
                        体面乡绅 硕果累累
                        编写于 最后由 编辑
                        #13

                        @imbiplaza-ASUS 说:

                        @Thanaots

                        提示词抽奖 + 视频生成抽奖

                        对,我抽奖了13个吧。只有一个勉勉强强,有个7分满意度。但是不想多试了,就定了。

                        1 条回复 最后回复
                        0

                        你好!看起来您对这段对话很感兴趣,但您还没有一个账号。

                        厌倦了每次访问都刷到同样的帖子?您注册账号后,您下次访问时都将自动回到上次浏览的位置,并可选择接收新回复的通知(通过电子邮件或推送通知)。您还可以收藏帖子、为帖子点赞,以此向其他社区成员表达您的感谢。

                        有了你的建议,这篇帖子会更精彩哦 💗

                        注册 登录
                        回复
                        • 在新帖中回复
                        登录后回复
                        • 从旧到新
                        • 从新到旧
                        • 最多赞同


                        • 登录

                        • 登录或注册以进行搜索。
                        • 第一个帖子
                          最后一个帖子
                        0
                        • 版块
                        • 最新
                        • 标签
                        • 热门
                        • 用户
                        • 群组