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. 【H3剑心比赛】RTX 5090 32GB,DDR5,3分56秒出片

【H3剑心比赛】RTX 5090 32GB,DDR5,3分56秒出片

已定时 已固定 已锁定 已移动 优秀 AI音视频画图
rtx5090minimax视频生成
9 帖子 4 发布者 354 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。
  • tester12340T
    tester12340T
    tester12340
    体面乡绅 硕果累累
    编写于 最后由 tester12340 编辑
    #1

    参赛:RTX 5090 32GB,MiniMax H3 参考图转视频(R2V),1344×768,12 秒,2D 和 3D 各一个

    油管链接

    • 3D:https://youtu.be/9rTUT9YIPl8
    • 2D:https://youtu.be/vaGFURkGrRQ

    一、硬件与环境

    • 显卡:NVIDIA GeForce RTX 5090 32GB,内存 64GB DDR5,Windows
    • ComfyUI 0.45.2,PyTorch 2.12.1+cu130,Python 3.13
    • 加速手段:SageAttention(启动参数 --use-sage-attention)、ComfyUI 动态显存加载(dynamic VRAM)、INT8 ConvRot 量化权重、8 步 Turbo LoRA

    二、模型与权重

    • 扩散模型:minimax_h3_ref2va_pruned_int8_convrot.safetensors
    • 加速 LoRA:minimax_h3_ref2v_turbo_8step_v1.0_768p_comfyui_bf16.safetensors(强度 1.0,8 步)
    • 文本编码器:qwen3vl_32b_heretic_minimax_h3_nvfp4.safetensors(nvfp4)
    • VAE:minimax_h3_video_vae_fp16 和 minimax_h3_audio_vae_fp32

    三、工作流

    • 官方模板 video_minimax_h3_r2v 的节点结构,核心节点是 MiniMaxH3ReferenceToVideo
    • 6 张参考图按顺序输入(剑心 3 张,志志雄 3 张),ref_image_size = match
    • 提示词使用帖子里的统一提示词,只替换风格那一句(3D CGI animation 或 2D traditional anime)
    • 采样器 res_multistep,调度器 simple,8 步,BasicGuider,不使用 CFG,种子 20261114
    • 分辨率 1344×768,帧数 294 帧,24fps,时长 12.25 秒,h264 加 aac 音频
    • 帧数说明:H3 的帧数必须满足 17k+5,12 秒(288 帧)会被向上取整到 294 帧
    • 工作流 JSON 见附件(API 格式)

    四、运行时间

    • 2D:Prompt executed in 303.33 秒(模型已预热,采样 8 步用时 3 分 56 秒,约 29.6 秒/步)
    • 3D:Prompt executed in 00:10:11(冷启动,包含 32B 文本编码器、DiT 和 VAE 的首次加载,采样 8 步用时 4 分 04 秒,约 30.6 秒/步)
    • 运行日志截图见附件

    五、说明

    • 这次用了 8 步 Turbo LoRA。
    1 条回复 最后回复
    0
    • tester12340T
      tester12340T
      tester12340
      体面乡绅 硕果累累
      编写于 最后由 编辑
      #2

      2d

      {
       "1": {
        "class_type": "UNETLoader",
        "inputs": {
         "unet_name": "minimax_h3_ref2va_pruned_int8_convrot.safetensors",
         "weight_dtype": "default"
        }
       },
       "2": {
        "class_type": "CLIPLoader",
        "inputs": {
         "clip_name": "qwen3vl_32b_heretic_minimax_h3_nvfp4.safetensors",
         "type": "minimax",
         "device": "default"
        }
       },
       "3": {
        "class_type": "VAELoader",
        "inputs": {
         "vae_name": "minimax_h3_video_vae_fp16.safetensors"
        }
       },
       "4": {
        "class_type": "VAELoader",
        "inputs": {
         "vae_name": "minimax_h3_audio_vae_fp32.safetensors"
        }
       },
       "5": {
        "class_type": "LoraLoaderModelOnly",
        "inputs": {
         "lora_name": "minimax_h3_ref2v_turbo_8step_v1.0_768p_comfyui_bf16.safetensors",
         "strength_model": 1,
         "model": [
          "1",
          0
         ]
        }
       },
       "20": {
        "class_type": "LoadImage",
        "inputs": {
         "image": "contest_kenshin1.jpeg"
        }
       },
       "21": {
        "class_type": "LoadImage",
        "inputs": {
         "image": "contest_kenshin2.jpeg"
        }
       },
       "22": {
        "class_type": "LoadImage",
        "inputs": {
         "image": "contest_kenshin3.jpeg"
        }
       },
       "23": {
        "class_type": "LoadImage",
        "inputs": {
         "image": "contest_shishio1.jpeg"
        }
       },
       "24": {
        "class_type": "LoadImage",
        "inputs": {
         "image": "contest_shishio2.jpeg"
        }
       },
       "25": {
        "class_type": "LoadImage",
        "inputs": {
         "image": "contest_shishio3.jpeg"
        }
       },
       "30": {
        "class_type": "MiniMaxH3ReferenceToVideo",
        "inputs": {
         "clip": [
          "2",
          0
         ],
         "vae": [
          "3",
          0
         ],
         "audio_vae": [
          "4",
          0
         ],
         "prompt": "(Style: 2D traditional anime) Masterpiece, high quality, cinematic lighting.\nSubject: A fierce battle sequence featuring two characters based on the reference images�a red-clad swordsman (Jianxin) and a purple-robed warrior (Zhizhixiong).\nAction: The camera begins with a dynamic low-angle tracking shot. The swordsman in red dashes forward with a long sword, leaving a glowing trail, while the warrior in purple stands firm and executes a sweeping counter-attack with a flaming effect. They clash in the center of the frame, creating a shockwave of energy and sparks. The camera rapidly orbits around them as the fiery battle intensifies, concluding with the red-clad swordsman landing gracefully while the purple warrior points his weapon forward, both facing the camera.\nEnvironment: Dramatic background with embers floating in the air, high contrast lighting.",
         "width": 1344,
         "height": 768,
         "length": 294,
         "ref_image_size": "match",
         "ref_images.ref_image_0": [
          "20",
          0
         ],
         "ref_images.ref_image_1": [
          "21",
          0
         ],
         "ref_images.ref_image_2": [
          "22",
          0
         ],
         "ref_images.ref_image_3": [
          "23",
          0
         ],
         "ref_images.ref_image_4": [
          "24",
          0
         ],
         "ref_images.ref_image_5": [
          "25",
          0
         ]
        }
       },
       "31": {
        "class_type": "KSamplerSelect",
        "inputs": {
         "sampler_name": "res_multistep"
        }
       },
       "32": {
        "class_type": "BasicScheduler",
        "inputs": {
         "model": [
          "5",
          0
         ],
         "scheduler": "simple",
         "steps": 8,
         "denoise": 1
        }
       },
       "33": {
        "class_type": "BasicGuider",
        "inputs": {
         "model": [
          "5",
          0
         ],
         "conditioning": [
          "30",
          0
         ]
        }
       },
       "34": {
        "class_type": "RandomNoise",
        "inputs": {
         "noise_seed": 20261114
        }
       },
       "35": {
        "class_type": "SamplerCustomAdvanced",
        "inputs": {
         "noise": [
          "34",
          0
         ],
         "guider": [
          "33",
          0
         ],
         "sampler": [
          "31",
          0
         ],
         "sigmas": [
          "32",
          0
         ],
         "latent_image": [
          "30",
          1
         ]
        }
       },
       "36": {
        "class_type": "VAEDecode",
        "inputs": {
         "samples": [
          "35",
          0
         ],
         "vae": [
          "3",
          0
         ]
        }
       },
       "37": {
        "class_type": "VAEDecodeAudio",
        "inputs": {
         "samples": [
          "35",
          0
         ],
         "vae": [
          "4",
          0
         ]
        }
       },
       "38": {
        "class_type": "CreateVideo",
        "inputs": {
         "images": [
          "36",
          0
         ],
         "audio": [
          "37",
          0
         ],
         "fps": 24
        }
       },
       "39": {
        "class_type": "SaveVideo",
        "inputs": {
         "video": [
          "38",
          0
         ],
         "filename_prefix": "contest/h3_2d_turbo",
         "format": "mp4",
         "codec": "h264"
        }
       }
      }
      

      3d

      {
       "1": {
        "class_type": "UNETLoader",
        "inputs": {
         "unet_name": "minimax_h3_ref2va_pruned_int8_convrot.safetensors",
         "weight_dtype": "default"
        }
       },
       "2": {
        "class_type": "CLIPLoader",
        "inputs": {
         "clip_name": "qwen3vl_32b_heretic_minimax_h3_nvfp4.safetensors",
         "type": "minimax",
         "device": "default"
        }
       },
       "3": {
        "class_type": "VAELoader",
        "inputs": {
         "vae_name": "minimax_h3_video_vae_fp16.safetensors"
        }
       },
       "4": {
        "class_type": "VAELoader",
        "inputs": {
         "vae_name": "minimax_h3_audio_vae_fp32.safetensors"
        }
       },
       "5": {
        "class_type": "LoraLoaderModelOnly",
        "inputs": {
         "lora_name": "minimax_h3_ref2v_turbo_8step_v1.0_768p_comfyui_bf16.safetensors",
         "strength_model": 1,
         "model": [
          "1",
          0
         ]
        }
       },
       "20": {
        "class_type": "LoadImage",
        "inputs": {
         "image": "contest_kenshin1.jpeg"
        }
       },
       "21": {
        "class_type": "LoadImage",
        "inputs": {
         "image": "contest_kenshin2.jpeg"
        }
       },
       "22": {
        "class_type": "LoadImage",
        "inputs": {
         "image": "contest_kenshin3.jpeg"
        }
       },
       "23": {
        "class_type": "LoadImage",
        "inputs": {
         "image": "contest_shishio1.jpeg"
        }
       },
       "24": {
        "class_type": "LoadImage",
        "inputs": {
         "image": "contest_shishio2.jpeg"
        }
       },
       "25": {
        "class_type": "LoadImage",
        "inputs": {
         "image": "contest_shishio3.jpeg"
        }
       },
       "30": {
        "class_type": "MiniMaxH3ReferenceToVideo",
        "inputs": {
         "clip": [
          "2",
          0
         ],
         "vae": [
          "3",
          0
         ],
         "audio_vae": [
          "4",
          0
         ],
         "prompt": "(Style: 3D CGI animation) Masterpiece, high quality, cinematic lighting.\nSubject: A fierce battle sequence featuring two characters based on the reference images�a red-clad swordsman (Jianxin) and a purple-robed warrior (Zhizhixiong).\nAction: The camera begins with a dynamic low-angle tracking shot. The swordsman in red dashes forward with a long sword, leaving a glowing trail, while the warrior in purple stands firm and executes a sweeping counter-attack with a flaming effect. They clash in the center of the frame, creating a shockwave of energy and sparks. The camera rapidly orbits around them as the fiery battle intensifies, concluding with the red-clad swordsman landing gracefully while the purple warrior points his weapon forward, both facing the camera.\nEnvironment: Dramatic background with embers floating in the air, high contrast lighting.",
         "width": 1344,
         "height": 768,
         "length": 294,
         "ref_image_size": "match",
         "ref_images.ref_image_0": [
          "20",
          0
         ],
         "ref_images.ref_image_1": [
          "21",
          0
         ],
         "ref_images.ref_image_2": [
          "22",
          0
         ],
         "ref_images.ref_image_3": [
          "23",
          0
         ],
         "ref_images.ref_image_4": [
          "24",
          0
         ],
         "ref_images.ref_image_5": [
          "25",
          0
         ]
        }
       },
       "31": {
        "class_type": "KSamplerSelect",
        "inputs": {
         "sampler_name": "res_multistep"
        }
       },
       "32": {
        "class_type": "BasicScheduler",
        "inputs": {
         "model": [
          "5",
          0
         ],
         "scheduler": "simple",
         "steps": 8,
         "denoise": 1
        }
       },
       "33": {
        "class_type": "BasicGuider",
        "inputs": {
         "model": [
          "5",
          0
         ],
         "conditioning": [
          "30",
          0
         ]
        }
       },
       "34": {
        "class_type": "RandomNoise",
        "inputs": {
         "noise_seed": 20261114
        }
       },
       "35": {
        "class_type": "SamplerCustomAdvanced",
        "inputs": {
         "noise": [
          "34",
          0
         ],
         "guider": [
          "33",
          0
         ],
         "sampler": [
          "31",
          0
         ],
         "sigmas": [
          "32",
          0
         ],
         "latent_image": [
          "30",
          1
         ]
        }
       },
       "36": {
        "class_type": "VAEDecode",
        "inputs": {
         "samples": [
          "35",
          0
         ],
         "vae": [
          "3",
          0
         ]
        }
       },
       "37": {
        "class_type": "VAEDecodeAudio",
        "inputs": {
         "samples": [
          "35",
          0
         ],
         "vae": [
          "4",
          0
         ]
        }
       },
       "38": {
        "class_type": "CreateVideo",
        "inputs": {
         "images": [
          "36",
          0
         ],
         "audio": [
          "37",
          0
         ],
         "fps": 24
        }
       },
       "39": {
        "class_type": "SaveVideo",
        "inputs": {
         "video": [
          "38",
          0
         ],
         "filename_prefix": "contest/h3_3d_turbo",
         "format": "mp4",
         "codec": "h264"
        }
       }
      }
      
      1 条回复 最后回复
      0
      • terryT
        terryT
        terry
        易有太极
        编写于 最后由 编辑
        #3

        通过初审,正式参赛,奖励积分20,比赛结束后统一结算!
        但是说实话:剑心3D识别不行。

        油管:https://www.youtube.com/@抡锤者

        tester12340T 1 条回复 最后回复
        0
        • XiaoteX
          XiaoteX
          Xiaote
          易有太极
          编写于 最后由 编辑
          #4

          ✅ 版主审定:本帖设为优秀,作者 +3 积分奖励,继续加油!

          奖励凭证:优秀 +3 分 · 编号 e3075-4234-1

          老特的AI助手,DeepSeek Flash驱动,没回你是因为被限速了~除了管理员不能私信我,别浪费时间~尝试多了被封号就不好了~

          1 条回复 最后回复
          0
          • terryT terry

            通过初审,正式参赛,奖励积分20,比赛结束后统一结算!
            但是说实话:剑心3D识别不行。

            tester12340T
            tester12340T
            tester12340
            体面乡绅 硕果累累
            编写于 最后由 编辑
            #5

            @terry 感谢! 3d识别不行 是没把3d的内容识别出来吗?

            Max WangM 1 条回复 最后回复
            0
            • tester12340T tester12340

              @terry 感谢! 3d识别不行 是没把3d的内容识别出来吗?

              Max WangM
              Max WangM
              Max Wang
              体面乡绅
              编写于 最后由 编辑
              #6

              @tester12340 应该是说这个3d视频的效果不够立体,种子方向过于偏向参考图里的2d立绘了

              terryT 1 条回复 最后回复
              1
              • Max WangM Max Wang

                @tester12340 应该是说这个3d视频的效果不够立体,种子方向过于偏向参考图里的2d立绘了

                terryT
                terryT
                terry
                易有太极
                编写于 最后由 terry 编辑
                #7

                @Max-Wang
                @tester12340
                是的。

                应该是说这个3d视频的效果不够立体,种子方向过于偏向参考图里的2d立绘了
                

                油管:https://www.youtube.com/@抡锤者

                1 条回复 最后回复
                0
                • tester12340T
                  tester12340T
                  tester12340
                  体面乡绅 硕果累累
                  编写于 最后由 编辑
                  #8

                  @terry @max-wang 原来如此。

                  1 条回复 最后回复
                  0
                  • ,tester12340T tester12340 引用了 此主题
                  • tester12340T
                    tester12340T
                    tester12340
                    体面乡绅 硕果累累
                    编写于 最后由 编辑
                    #9
                    此主題已被删除!
                    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