1
00:00:00,000 --> 00:00:29,280
这个频道做的是把几百年前的城市用AI一帧一帧的重建出来。1600年的伦敦长什么样,1700年的费城长什么样,频道达到500多万次的观看,变现至少在2万美金以上。很多人看到这个时长被炫退了,毕竟感觉每一张图都要一帧一帧的生成,还要写提示词。但这个是我用一个AI一键生成的几百个分镜,讲的是罗马的历史,每一个分镜都跟文案强相关,每张图质量都非常高。

2
00:00:29,280 --> 00:00:30,320
大家可以看到

3
00:00:30,320 --> 00:00:31,420
这个时间间隔

4
00:00:31,420 --> 00:00:32,880
总共用了两个多小时

5
00:00:32,880 --> 00:00:34,480
AI自动排队生成

6
00:00:34,480 --> 00:00:35,900
还有自动生成的视频

7
00:00:35,900 --> 00:00:37,540
方法我们等一下会说

8
00:00:37,540 --> 00:00:38,760
我们先来看一小段

9
00:00:38,760 --> 00:00:40,220
我用差不多10分钟左右

10
00:00:40,220 --> 00:00:41,160
做出来的成片

11
00:00:59,280 --> 00:01:08,460
这个视频的成片大概半个多小时

12
00:01:08,460 --> 00:01:10,600
AI自动画跑了两个小时左右

13
00:01:10,600 --> 00:01:13,140
具体的制作方法我们等一下也会说

14
00:01:13,140 --> 00:01:15,840
如果你觉得做这种频道还是用语言壁垒

15
00:01:15,840 --> 00:01:17,480
对于做英文频道打触

16
00:01:17,480 --> 00:01:19,400
或者说对于做长视频打触

17
00:01:19,400 --> 00:01:20,700
我们来看这个频道

18
00:01:20,700 --> 00:01:23,040
做的是吉普利风格的怀旧动画

19
00:01:23,040 --> 00:01:25,000
总共只发布了36个视频

20
00:01:25,000 --> 00:01:27,180
但是已经得到了800万次的观看

21
00:01:27,180 --> 00:01:31,100
就算是拿最低的2500美金每百万播放的RPM来算

22
00:01:31,100 --> 00:01:32,740
也已经拿到了2万美金

23
00:01:32,740 --> 00:01:35,080
视频全部都是90年代的乡村

24
00:01:35,080 --> 00:01:36,300
雨季 火车

25
00:01:36,300 --> 00:01:38,180
一家人在土造前面做饭

26
00:01:38,180 --> 00:01:39,680
没有旁白 没对话

27
00:01:39,680 --> 00:01:41,140
没有任何的语言壁垒

28
00:01:41,140 --> 00:01:43,220
完全是宫崎骏风格的动画

29
00:01:43,220 --> 00:01:44,820
只有音效和BGM

30
00:01:44,820 --> 00:01:46,540
营造出来很祥和的感觉

31
00:01:46,540 --> 00:01:48,980
如果你还是觉得做视频的压力太大

32
00:01:48,980 --> 00:01:50,160
或者费用太高

33
00:01:50,160 --> 00:01:51,720
我们来看第三个频道

34
00:01:51,720 --> 00:01:53,500
做的是远古人类的故事

35
00:01:53,500 --> 00:01:55,920
它的制作难度是三个里面最低的

36
00:01:55,920 --> 00:01:57,320
它甚至连动画都没有

37
00:01:57,320 --> 00:02:00,120
主体就是一张一张AI生成的图片在切

38
00:02:00,120 --> 00:02:00,940
换句话说

39
00:02:00,940 --> 00:02:02,820
连生视频的成本都不需要付

40
00:02:02,820 --> 00:02:05,460
而这些图片成本在现在真的很低

41
00:02:05,460 --> 00:02:07,340
这三个赛道的共同点就是

42
00:02:07,340 --> 00:02:10,840
从素材到成片全部都可以用AI跑出来

43
00:02:10,840 --> 00:02:12,340
不需要买素材出镜

44
00:02:12,340 --> 00:02:13,460
更不需要拍摄

45
00:02:13,460 --> 00:02:14,880
现在这期视频的主题呢

46
00:02:14,880 --> 00:02:16,940
不仅仅是教你怎么做AI视频

47
00:02:16,940 --> 00:02:18,440
更多的是怎么样学成

48
00:02:18,440 --> 00:02:19,720
怎么样去提升效率

49
00:02:19,720 --> 00:02:21,540
因为真正卡住大部分人的

50
00:02:21,540 --> 00:02:23,520
并不是AI会不会用这件事

51
00:02:23,520 --> 00:02:24,920
毕竟大家现在都知道

52
00:02:24,920 --> 00:02:26,960
一句提示词就可以生成一张图

53
00:02:26,960 --> 00:02:27,960
但是这种视频

54
00:02:27,960 --> 00:02:29,400
不仅仅是要生图

55
00:02:29,400 --> 00:02:30,260
还需要配音

56
00:02:30,260 --> 00:02:30,940
生视频

57
00:02:30,940 --> 00:02:32,040
剪辑一大堆

58
00:02:32,040 --> 00:02:34,100
一堆工具切换下来非常的碎

59
00:02:34,100 --> 00:02:35,180
你做一百个分镜

60
00:02:35,180 --> 00:02:37,660
可能就要在这些工具之间切上百次

61
00:02:37,660 --> 00:02:39,740
看到网上所谓的AI视频工作流

62
00:02:39,740 --> 00:02:40,600
也一头雾水

63
00:02:40,600 --> 00:02:42,480
所以今天这一视频的主要目的

64
00:02:42,480 --> 00:02:43,640
是手把手的

65
00:02:43,640 --> 00:02:45,780
带着大家从熟悉到熟练

66
00:02:45,780 --> 00:02:46,920
掌握AI视频

67
00:02:46,920 --> 00:02:48,220
以及自动化流程

68
00:02:48,220 --> 00:02:50,460
一键就能够调用你所有需要的工具

69
00:02:50,460 --> 00:02:51,560
并且能够变现

70
00:02:51,560 --> 00:02:53,000
也包括如何做选题

71
00:02:53,000 --> 00:02:53,600
选赛道

72
00:02:53,600 --> 00:02:54,220
Hello大家好

73
00:02:54,220 --> 00:02:54,760
我是安然

74
00:02:54,760 --> 00:02:56,280
我这个频道是专门分享

75
00:02:56,280 --> 00:02:57,720
普通人在这个AI时代

76
00:02:57,720 --> 00:02:59,180
怎么搭建自己的一人公司

77
00:02:59,180 --> 00:03:00,580
自己的商业模式的

78
00:03:00,580 --> 00:03:02,460
如果你也想要抓住这一波红利

79
00:03:02,460 --> 00:03:03,540
别忘了订阅我

80
00:03:03,540 --> 00:03:04,580
按下旁边的小铃铛

81
00:03:04,580 --> 00:03:05,160
点一个赞

82
00:03:05,160 --> 00:03:06,400
这样你就不会错过

83
00:03:06,400 --> 00:03:08,240
我们后续任何有价值的内容啦

84
00:03:08,240 --> 00:03:17,180
今天一整套流程

85
00:03:17,180 --> 00:03:19,760
我们全用一个PixWars平台跑完

86
00:03:19,760 --> 00:03:21,900
编接我放在描述栏里的第一条

87
00:03:21,900 --> 00:03:23,580
你可以把它理解成一个

88
00:03:23,580 --> 00:03:26,960
把声图 声视频 配音 配乐 角色一致性

89
00:03:26,960 --> 00:03:29,500
批量任务全部都能装进去的工作台

90
00:03:29,500 --> 00:03:32,300
它自己有一个模型叫做Pixworth V6

91
00:03:32,300 --> 00:03:34,520
也有SynDesk Minimax H3

92
00:03:34,520 --> 00:03:37,700
可灵 Vivo这些模型都是在这里并排排着的

93
00:03:37,700 --> 00:03:40,580
你并不需要为每一个模型单独的注册账号

94
00:03:40,580 --> 00:03:42,420
单独充值在Pixworth

95
00:03:42,420 --> 00:03:44,500
你可以根据自己想要的结果

96
00:03:44,500 --> 00:03:46,100
随时选择不同的工具

97
00:03:46,100 --> 00:03:46,820
可能你会说

98
00:03:46,820 --> 00:03:49,280
这个看起来跟其他的AI工作台产品

99
00:03:49,280 --> 00:03:50,640
没有什么太大区别呀

100
00:03:50,640 --> 00:03:52,300
但实际上它最大的优势

101
00:03:52,300 --> 00:03:53,360
一方面是效率

102
00:03:53,360 --> 00:03:54,840
另外一方面是成本

103
00:03:54,840 --> 00:03:56,740
做AI视频的最大卡点

104
00:03:56,740 --> 00:03:57,900
某种意义上来说

105
00:03:57,900 --> 00:03:59,420
并不是在于技术本身

106
00:03:59,420 --> 00:04:01,600
你看这种比较精致的画面和转场

107
00:04:01,600 --> 00:04:04,080
我们需要用Synthes H3来做

108
00:04:04,080 --> 00:04:06,120
但一个视频少则是几分钟

109
00:04:06,120 --> 00:04:07,140
长则几小时

110
00:04:07,140 --> 00:04:09,480
其他大量画面如果都用这个模型

111
00:04:09,480 --> 00:04:11,020
在成本上根本吃不消

112
00:04:11,020 --> 00:04:13,440
网上有很多的博主教你怎么做AI视频

113
00:04:13,440 --> 00:04:15,360
但是很少有人提到成本问题

114
00:04:15,360 --> 00:04:16,940
你的提示词写的再酷炫

115
00:04:16,940 --> 00:04:18,760
想要表达的东西再丰富

116
00:04:18,760 --> 00:04:20,640
但如果AI视频无法变现

117
00:04:20,640 --> 00:04:22,580
或者说前期的成本非常高

118
00:04:22,580 --> 00:04:24,540
那条视频就要花好几百美金

119
00:04:24,540 --> 00:04:25,780
那有什么意义呢

120
00:04:25,780 --> 00:04:27,840
它的Pixos V6视频模型

121
00:04:27,840 --> 00:04:31,180
我们完全可以跟Synthes和MiniMax H3搭配来做

122
00:04:31,180 --> 00:04:32,960
尽管说效果不如Synthes

123
00:04:32,960 --> 00:04:35,400
但又不是所有的画面都是大场面

124
00:04:35,400 --> 00:04:36,440
完全够用了

125
00:04:36,440 --> 00:04:38,980
而成本是Synthes的将近十分之一

126
00:04:38,980 --> 00:04:40,600
我们先来说第一个赛道

127
00:04:40,600 --> 00:04:42,000
历史类纪录片

128
00:04:42,000 --> 00:04:44,220
这是我感觉天文版最高的一个

129
00:04:44,220 --> 00:04:45,580
你看这个频道的名字

130
00:04:45,580 --> 00:04:47,120
直接就叫Our历史

131
00:04:47,120 --> 00:04:48,880
一看就是我们国人做出来的

132
00:04:48,880 --> 00:04:51,080
这三个视频就干到了百万播放

133
00:04:51,080 --> 00:04:52,940
这个频道做的是朝代讲解

134
00:04:52,940 --> 00:04:54,220
它做的事情很简单

135
00:04:54,220 --> 00:04:56,180
把历史上某个年代的样子

136
00:04:56,180 --> 00:04:57,900
比如说这个视频只是唐朝

137
00:04:57,900 --> 00:04:59,920
用AI一帧一帧的还原出来

138
00:04:59,920 --> 00:05:02,000
配上舒缓的解说或者纯音乐

139
00:05:02,000 --> 00:05:04,380
做成十几分钟甚至更长的视频

140
00:05:04,380 --> 00:05:07,560
这个赛道所有的素材全部都是AI生成的

141
00:05:07,560 --> 00:05:08,940
你不需要去买素材裤

142
00:05:08,940 --> 00:05:09,680
安心版权

143
00:05:09,680 --> 00:05:11,540
你要做1600年的伦敦

144
00:05:11,540 --> 00:05:13,200
或者做明朝的讲解

145
00:05:13,200 --> 00:05:16,100
那个年代的画面本来就没有任何影像资料

146
00:05:16,100 --> 00:05:18,260
全世界谁也拿不出来真实素材

147
00:05:18,260 --> 00:05:20,620
所以AI生成在这里不是替代笔

148
00:05:20,620 --> 00:05:21,760
它就是唯一的方案

149
00:05:21,760 --> 00:05:24,160
像开头展示的西方国家更可以做

150
00:05:24,160 --> 00:05:25,300
比如这个频道就是

151
00:05:25,300 --> 00:05:27,220
三个视频就干到了百万播放

152
00:05:27,220 --> 00:05:28,980
讲的也是西方国家的历史

153
00:05:28,980 --> 00:05:31,540
甚至说如果你对于国家历史不感兴趣

154
00:05:31,540 --> 00:05:33,000
可以像这个账号一样

155
00:05:33,000 --> 00:05:34,420
讲各种食物的历史

156
00:05:34,420 --> 00:05:37,220
50多个视频就带来了300多万的播放

157
00:05:37,220 --> 00:05:40,000
这类视频观众的停留时间天然就长

158
00:05:40,000 --> 00:05:42,140
它不是只看15秒刷一下就走的

159
00:05:42,140 --> 00:05:44,280
很多人是当做纪录片放着看的

160
00:05:44,280 --> 00:05:45,340
停留时间长

161
00:05:45,340 --> 00:05:48,020
意味着一条视频能够差更多的广告位

162
00:05:48,020 --> 00:05:49,640
收益结构完全不一样

163
00:05:49,640 --> 00:05:53,100
这类内容在算法眼里属于有信息增量的内容

164
00:05:53,100 --> 00:05:56,400
我之前专门做过一期讲YouTube打击什么内容的视频

165
00:05:56,400 --> 00:05:57,760
结论其实就一句话

166
00:05:57,760 --> 00:06:00,180
YouTube打击的是重复性和垃圾内容

167
00:06:00,180 --> 00:06:01,480
不是打击AI

168
00:06:01,480 --> 00:06:03,920
只要你的内容里面有足够多的创意

169
00:06:03,920 --> 00:06:05,720
有明显人类参与的判断

170
00:06:05,720 --> 00:06:07,680
那该怎么赚钱还是怎么赚钱

171
00:06:07,680 --> 00:06:09,060
我们下来说选题

172
00:06:09,060 --> 00:06:10,900
无论你是想要做任何赛道

173
00:06:10,900 --> 00:06:13,200
只要是你想要靠着流量赚钱

174
00:06:13,200 --> 00:06:14,920
都可以打开你常用的AI

175
00:06:14,920 --> 00:06:15,740
进键对话

176
00:06:15,740 --> 00:06:17,340
粘贴下面这段提示词

177
00:06:17,340 --> 00:06:19,420
这个选题提示词我已经调好了

178
00:06:19,420 --> 00:06:20,580
你可以直接复制

179
00:06:20,580 --> 00:06:22,080
第一条和第二条最重要

180
00:06:22,080 --> 00:06:24,700
因为AI在历史题材上特别容易编

181
00:06:24,700 --> 00:06:27,560
你不给他一个不确定就跳过的应约数

182
00:06:27,560 --> 00:06:29,000
待会给你造出来一堆

183
00:06:29,000 --> 00:06:31,980
根本没有任何史料支撑的城市年代组合

184
00:06:31,980 --> 00:06:34,000
然后我们看到了这么多的选题

185
00:06:34,000 --> 00:06:35,960
可能你觉得选的困难正范了

186
00:06:35,960 --> 00:06:36,840
到底做哪一个

187
00:06:36,840 --> 00:06:38,280
感觉每一个都挺好

188
00:06:38,280 --> 00:06:40,080
而且AI说的有理有据

189
00:06:40,080 --> 00:06:41,280
选题方法筛选

190
00:06:41,280 --> 00:06:42,500
我给你一个特别简单的

191
00:06:42,500 --> 00:06:45,020
你先选一个自己感兴趣的城市和年代

192
00:06:45,020 --> 00:06:47,860
然后直接去YouTube搜这个组合的英文关键词

193
00:06:47,860 --> 00:06:48,720
搜完之后

194
00:06:48,720 --> 00:06:50,840
一定要选择这个筛选器

195
00:06:50,840 --> 00:06:52,820
把上传的日期改成本月

196
00:06:52,820 --> 00:06:54,260
如果最近一个月里面

197
00:06:54,260 --> 00:06:56,080
有新账号靠着这个城市

198
00:06:56,080 --> 00:06:57,440
跑出来了不错的播放

199
00:06:57,440 --> 00:06:58,980
说明观众的兴趣还在

200
00:06:58,980 --> 00:06:59,700
你可以进

201
00:06:59,700 --> 00:07:00,680
如果一个月内

202
00:07:00,680 --> 00:07:01,940
做这个选题的账号

203
00:07:01,940 --> 00:07:03,120
数据都非常的惨

204
00:07:03,120 --> 00:07:04,620
那要不然是太选了

205
00:07:04,620 --> 00:07:07,460
要不然是头部已经把这个城市做透了

206
00:07:07,460 --> 00:07:08,740
大家兴趣转移了

207
00:07:08,740 --> 00:07:09,460
亦或者是

208
00:07:09,460 --> 00:07:11,080
大家其实根本不感兴趣

209
00:07:11,080 --> 00:07:12,900
反过来我们也可以去注意

210
00:07:12,900 --> 00:07:14,020
如果一个月内

211
00:07:14,020 --> 00:07:15,420
几乎没人做这个城市

212
00:07:15,420 --> 00:07:18,060
但是同一个国家的其他城市在起量

213
00:07:18,060 --> 00:07:19,940
那这个有可能就是一个空位

214
00:07:19,940 --> 00:07:22,940
西方城市留下来的考据资料相对来说的多

215
00:07:22,940 --> 00:07:25,320
17世纪的伦敦有全景版画

216
00:07:25,320 --> 00:07:27,500
19世纪的巴黎有城市地图

217
00:07:27,500 --> 00:07:30,380
20世纪初的纽约有大量的照片档案

218
00:07:30,380 --> 00:07:33,640
如果你对于做西方的素材实在是不熟悉打触

219
00:07:33,640 --> 00:07:36,280
那也可以像之前展示的那类账号一样

220
00:07:36,280 --> 00:07:37,140
做中国的

221
00:07:37,140 --> 00:07:38,620
也有很多的画和史料

222
00:07:38,620 --> 00:07:40,060
这些东西你能查到

223
00:07:40,060 --> 00:07:41,900
AI生成出来就有据可依

224
00:07:41,900 --> 00:07:42,780
而不是瞎编

225
00:07:42,780 --> 00:07:45,620
我在这里先选的是美国革命战争的讲述

226
00:07:45,620 --> 00:07:48,640
因为那个时候的美国和今天几乎是两个国家

227
00:07:48,640 --> 00:07:50,820
而且很容易唤起美国人的好奇心

228
00:07:50,820 --> 00:07:52,120
吸引到美国的观众

229
00:07:52,120 --> 00:07:54,600
我们现在开始用PixWords生成视频

230
00:07:54,600 --> 00:07:56,860
先来说小白用法再来说进阶

231
00:07:56,860 --> 00:07:59,740
很多小白做视频是一次就生成一条

232
00:08:04,000 --> 00:08:14,000
這樣子做一條短視頻還行。如果做一條有200個分鏡的長視頻,一到第五個分鏡可能就開始亂了。哪個提示詞對應哪張圖,哪張是選中的,哪張是廢掉的。

233
00:08:14,000 --> 00:08:15,840
参考图传到第几张了

234
00:08:15,840 --> 00:08:16,880
Pixworks在这里

235
00:08:16,880 --> 00:08:18,940
也有常规的单张生成的窗口

236
00:08:18,940 --> 00:08:20,680
但对于我们的批量化作业

237
00:08:20,680 --> 00:08:21,700
我更推荐的是

238
00:08:21,700 --> 00:08:23,340
用Pixworks的无线画布

239
00:08:23,340 --> 00:08:24,640
大部分是一个生成框

240
00:08:24,640 --> 00:08:26,980
是一块把整个项目摊开的板子

241
00:08:26,980 --> 00:08:28,100
写在无线画布上

242
00:08:28,100 --> 00:08:30,120
你也可以一键调用所有的工具

243
00:08:30,120 --> 00:08:31,860
它的最小的单位叫做节点

244
00:08:31,860 --> 00:08:34,080
每个节点就是画布上的一张卡片

245
00:08:34,080 --> 00:08:34,960
一段文字

246
00:08:34,960 --> 00:08:36,140
一张参考图

247
00:08:36,140 --> 00:08:37,740
一张结构画的分镜表

248
00:08:37,740 --> 00:08:39,180
这些全部都是节点

249
00:08:39,180 --> 00:08:40,100
你的提示词

250
00:08:40,100 --> 00:08:40,680
图片

251
00:08:40,680 --> 00:08:41,320
视频

252
00:08:41,320 --> 00:08:43,500
都可以在这一张画布里面生成

253
00:08:43,500 --> 00:08:45,080
你把节点之间连起来

254
00:08:45,080 --> 00:08:47,320
前一个的输出就是后一个的输入

255
00:08:47,320 --> 00:08:48,860
角色图连到分信图

256
00:08:48,860 --> 00:08:50,080
分信图连到视频

257
00:08:50,080 --> 00:08:51,500
一条线串下来

258
00:08:51,500 --> 00:08:53,300
整个项目的结构是看得见的

259
00:08:53,300 --> 00:08:55,400
你甚至可以从文案方面开始出

260
00:08:55,400 --> 00:08:56,240
比如在这里

261
00:08:56,240 --> 00:08:57,400
我让他给我写一个

262
00:08:57,400 --> 00:08:59,740
讲述美国独立战争的英语口播文案

263
00:08:59,740 --> 00:09:01,080
这样子生成完之后

264
00:09:01,080 --> 00:09:03,540
我们就可以直接在无线画布上面操作

265
00:09:03,540 --> 00:09:04,640
很快就生成好了

266
00:09:04,640 --> 00:09:06,360
前面这里是全英文版

267
00:09:06,360 --> 00:09:09,000
然后我们可以去用中文来对照着看

268
00:09:09,000 --> 00:09:11,580
或者说如果你用其他的AI生成的文案

269
00:09:11,580 --> 00:09:12,620
复制过来也可以

270
00:09:12,620 --> 00:09:15,300
生存之后你可以选择让它自己生图

271
00:09:15,300 --> 00:09:18,040
也可以让它直接帮你生成一个分镜表出来

272
00:09:18,040 --> 00:09:19,140
直接给它提示词

273
00:09:19,140 --> 00:09:21,920
让它以5秒或者10秒钟为一个间隙

274
00:09:21,920 --> 00:09:22,880
生成分镜表

275
00:09:22,880 --> 00:09:25,600
这个分镜表需要包含所有的分镜提示词

276
00:09:25,600 --> 00:09:28,500
以及每一句每一个分镜里面需要的台词

277
00:09:28,500 --> 00:09:30,120
因为这个分镜表太长

278
00:09:30,120 --> 00:09:31,600
在网页里面不好展示

279
00:09:31,600 --> 00:09:35,080
我直接下载下来给大家看这个更完整的Excel版本

280
00:09:35,080 --> 00:09:37,040
你在无线画布里面的所有产出

281
00:09:37,040 --> 00:09:38,580
也都是可以单独导出的

282
00:09:38,580 --> 00:09:40,240
导出也是我们正常的格式

283
00:09:40,240 --> 00:09:42,200
这个里面包含了文案原文

284
00:09:42,200 --> 00:09:43,800
这一句话的中文翻译

285
00:09:43,800 --> 00:09:45,580
以及做这张图片的提示词

286
00:09:45,580 --> 00:09:47,500
所以接下来我们要做的事情

287
00:09:47,500 --> 00:09:50,300
实际上就是让AI照着这个英文提示词

288
00:09:50,300 --> 00:09:51,980
一张一张的生成出来

289
00:09:51,980 --> 00:09:54,940
然后如果我们觉得哪里或者哪一张有问题的话

290
00:09:54,940 --> 00:09:56,280
可以直接让AI来改

291
00:09:56,280 --> 00:09:58,740
最右边这里是生成视频的提示词

292
00:09:58,740 --> 00:09:59,540
也就是说

293
00:09:59,540 --> 00:10:01,460
当我们生成一张图片之后

294
00:10:01,460 --> 00:10:03,340
它怎么去转换成视频的格式

295
00:10:03,340 --> 00:10:05,980
并且所有的都效跟文案的原文相符

296
00:10:05,980 --> 00:10:08,420
我们可以看到一张分析表里面

297
00:10:08,420 --> 00:10:10,900
包含整个这一篇的大概300个提示词

298
00:10:10,900 --> 00:10:12,720
全都是AI一口气生成的

299
00:10:12,720 --> 00:10:14,120
这个里面每一个分镜

300
00:10:14,120 --> 00:10:15,320
包括中文翻译

301
00:10:15,320 --> 00:10:16,420
全部都准备好了

302
00:10:16,420 --> 00:10:19,260
这个工作量在人工时代是难以估计的

303
00:10:19,260 --> 00:10:22,700
但是现在Pixwords花大概十几二十分钟就能做好了

304
00:10:22,700 --> 00:10:23,840
那在一开始的时候

305
00:10:23,840 --> 00:10:25,160
我们肯定是要做实验

306
00:10:25,160 --> 00:10:27,520
看一看究竟我们想要什么样的风格

307
00:10:27,520 --> 00:10:29,600
再让AI去批量化的生成

308
00:10:29,600 --> 00:10:31,180
但是一开始的风格提示词

309
00:10:31,180 --> 00:10:32,420
可能我们又不会写

310
00:10:32,420 --> 00:10:33,260
所以在这里

311
00:10:33,260 --> 00:10:35,620
你可以先输入一个简单的提示词

312
00:10:35,620 --> 00:10:36,580
很短的话

313
00:10:36,580 --> 00:10:38,400
1600年的伦敦街道

314
00:10:38,400 --> 00:10:40,580
如果你就这个样子发给作徒的AI

315
00:10:40,580 --> 00:10:41,960
说实话正常的AI

316
00:10:41,960 --> 00:10:43,760
根本不知道你要什么光线

317
00:10:43,760 --> 00:10:44,480
什么季节

318
00:10:44,480 --> 00:10:45,160
什么视角

319
00:10:45,160 --> 00:10:46,400
但是我发送以后

320
00:10:46,400 --> 00:10:47,740
它会把这一句短句

321
00:10:47,740 --> 00:10:49,920
过写成一段结构完整的描述

322
00:10:49,920 --> 00:10:50,740
把主体

323
00:10:50,740 --> 00:10:51,380
动作

324
00:10:51,380 --> 00:10:51,920
环境

325
00:10:51,920 --> 00:10:52,540
镜头

326
00:10:52,540 --> 00:10:53,140
光线

327
00:10:53,140 --> 00:10:53,660
风格

328
00:10:53,660 --> 00:10:55,280
这几层全部都图上

329
00:10:55,280 --> 00:10:57,800
然后如果你想就用这个题字词来生图

330
00:10:57,800 --> 00:10:59,100
或者说你改了之后

331
00:10:59,100 --> 00:11:00,620
你可以直接加一个节点

332
00:11:00,620 --> 00:11:01,360
选择图片

333
00:11:01,360 --> 00:11:03,040
你不需要任何复制粘贴

334
00:11:03,040 --> 00:11:04,860
因为它已经有了上一步的结果

335
00:11:04,860 --> 00:11:06,140
你可以直接生成

336
00:11:06,140 --> 00:11:08,460
我们看到这个结果很快就生成出来了

337
00:11:08,460 --> 00:11:09,540
如果你觉得满意

338
00:11:09,540 --> 00:11:11,020
就在这里再加一个节点

339
00:11:11,020 --> 00:11:12,200
让它来生成视频

340
00:11:12,200 --> 00:11:13,220
我们看到在这里

341
00:11:13,220 --> 00:11:15,380
它自动的就把这张图作为了手针

342
00:11:15,380 --> 00:11:17,740
我的模型先选择Pixos V6

343
00:11:17,740 --> 00:11:19,520
我们也可以在这里选择时间

344
00:11:19,520 --> 00:11:20,860
玩法是非常多的

345
00:11:20,860 --> 00:11:23,180
包括它也可以专门作为转场生成

346
00:11:23,180 --> 00:11:24,780
大家也可以去探索一下

347
00:11:24,780 --> 00:11:26,960
我们看到这个视频很快就生成出来了

348
00:11:26,960 --> 00:11:28,780
而且在生成这个视频的期间

349
00:11:28,780 --> 00:11:30,420
其实你也可以去干其他事

350
00:11:30,420 --> 00:11:32,520
因为生成之后立马就会显示出来

351
00:11:32,520 --> 00:11:34,000
这个是为什么无线画布

352
00:11:34,000 --> 00:11:36,620
可以说无论是你是做短句还是慢句

353
00:11:36,620 --> 00:11:38,080
都不可或缺的一个工具

354
00:11:38,080 --> 00:11:39,740
然后我们来到关键的一步

355
00:11:39,740 --> 00:11:41,600
很多AI视频看起来廉价

356
00:11:41,600 --> 00:11:43,740
是因为它是一段一段拼出来的

357
00:11:43,740 --> 00:11:45,140
每五秒钟硬切一次

358
00:11:45,140 --> 00:11:47,640
观众的注意力每五秒钟被打断一次

359
00:11:47,640 --> 00:11:49,460
而真正好看的历史纪录片

360
00:11:49,460 --> 00:11:50,780
镜头是连着走的

361
00:11:50,780 --> 00:11:51,840
一路推进去

362
00:11:51,840 --> 00:11:54,100
观众跟着镜头在那个时代里面走

363
00:11:54,100 --> 00:11:54,880
大家可以看

364
00:11:54,880 --> 00:11:56,240
用同样一组图片

365
00:11:56,240 --> 00:11:58,740
这个是我用五秒分镜来切出来的

366
00:11:58,740 --> 00:12:01,780
Iran speaking peoples, including Medes and Persians,

367
00:12:01,780 --> 00:12:06,260
arived on the plateau in larger numbers during the second millennium BC.

368
00:12:06,260 --> 00:12:11,140
Iran eventually grew from an old expression for the realm of Iranian peoples.

369
00:12:11,140 --> 00:12:16,100
In this setting, the related word Aryan meant Iranian speaking communities.

370
00:12:16,100 --> 00:12:18,260
而这个是我用手伟真来做的。

371
00:12:18,260 --> 00:12:21,300
Iran speaking peoples, including Medes and Persians,

372
00:12:21,300 --> 00:12:25,700
arived on the plateau in larger numbers during the second millennium BC.

373
00:12:25,700 --> 00:12:30,300
以安国为以安国人类的古代代表

374
00:12:30,300 --> 00:12:31,440
在这种角度

375
00:12:31,440 --> 00:12:32,360
同样的词词

376
00:12:32,360 --> 00:12:34,240
你直观对比一下

377
00:12:34,240 --> 00:12:35,260
同样的素材

378
00:12:35,260 --> 00:12:37,580
右边就是有一种高级的感觉

379
00:12:37,580 --> 00:12:40,760
在这里我用的是minimax H3的transaction模式

380
00:12:40,760 --> 00:12:42,860
逻辑是给它一张开始的画面

381
00:12:42,860 --> 00:12:44,840
一张结束的画面作为首尾帧

382
00:12:44,840 --> 00:12:47,540
在无线画布上用节点来连接起来

383
00:12:47,540 --> 00:12:49,640
模型选择minimax H3

384
00:12:49,640 --> 00:12:52,540
我在这里直接选择15秒120P

385
00:12:52,540 --> 00:12:54,660
再加上文字说明中间怎么变

386
00:12:54,660 --> 00:12:56,520
它把中间的过程给补出来

387
00:12:56,520 --> 00:12:58,400
让它输出一段完整的运镜

388
00:12:58,400 --> 00:12:59,280
而不是硬切

389
00:12:59,280 --> 00:13:00,600
这个运镜的提示词

390
00:13:00,600 --> 00:13:01,900
也可以让AI来做

391
00:13:01,900 --> 00:13:03,100
连接这两张图

392
00:13:03,100 --> 00:13:05,080
让AI帮你优化提示词就可以

393
00:13:05,080 --> 00:13:06,880
你也可以尝试同一个镜头

394
00:13:06,880 --> 00:13:08,480
跑多个模型做对比

395
00:13:08,480 --> 00:13:10,500
比如说Synthense 2.0 2.5

396
00:13:10,500 --> 00:13:11,800
还有MiniMax H3

397
00:13:11,800 --> 00:13:14,100
在长视频里面的表现是不一样的

398
00:13:14,100 --> 00:13:15,100
你可以去做实验

399
00:13:15,100 --> 00:13:17,460
看到底哪一个更适合你这个镜头

400
00:13:17,460 --> 00:13:19,360
说实话不去试是不知道的

401
00:13:19,360 --> 00:13:22,640
PixWars也可以一键调用Synthense 2.5和2.0

402
00:13:22,640 --> 00:13:23,760
模型都非常全

403
00:13:23,760 --> 00:13:24,600
在画布上

404
00:13:24,600 --> 00:13:26,840
你可以从同一个节点拉出来两条线

405
00:13:26,840 --> 00:13:27,860
接到两个模型

406
00:13:27,860 --> 00:13:28,980
输入完全一致

407
00:13:28,980 --> 00:13:30,000
同样的提示词

408
00:13:30,000 --> 00:13:30,740
参考图

409
00:13:30,740 --> 00:13:32,120
还有同样的手为针

410
00:13:32,120 --> 00:13:33,940
然后把结果并排着放着笔

411
00:13:33,940 --> 00:13:35,480
而且它会在你跑之前

412
00:13:35,480 --> 00:13:36,540
告诉你这一次

413
00:13:36,540 --> 00:13:38,080
大概销告多少额度

414
00:13:38,080 --> 00:13:39,320
你可以先算一笔账

415
00:13:39,320 --> 00:13:41,040
再决定要不要跑大批量

416
00:13:41,040 --> 00:13:42,820
我的习惯是拿一个镜头

417
00:13:42,820 --> 00:13:44,680
跑三个模型选出最合适的

418
00:13:44,680 --> 00:13:46,060
剩下七八十个镜头

419
00:13:46,060 --> 00:13:47,680
全部都用这个模型来跑

420
00:13:47,680 --> 00:13:48,600
你可能会说

421
00:13:48,600 --> 00:13:51,280
好那我知道怎么样去升视频了

422
00:13:51,280 --> 00:13:52,540
但是就像你说的

423
00:13:52,540 --> 00:13:54,000
所有镜头都用Synthes

424
00:13:54,000 --> 00:13:55,000
是吃不消的

425
00:13:55,000 --> 00:13:55,840
对没错的

426
00:13:55,840 --> 00:13:56,780
所以说在这里

427
00:13:56,780 --> 00:13:58,960
我们说一下怎么样能够省下来成本

428
00:13:58,960 --> 00:14:00,620
其实对于YouTube来说

429
00:14:00,620 --> 00:14:02,900
最最重要的是前面30秒的留存

430
00:14:02,900 --> 00:14:03,800
我的建议是

431
00:14:03,800 --> 00:14:06,980
开头可以全部用Synthense或者MiniMax H3来做

432
00:14:06,980 --> 00:14:08,640
后面穿插着来就行了

433
00:14:08,640 --> 00:14:10,300
其实后面大部分的信头

434
00:14:10,300 --> 00:14:12,020
我们可以用Pixworth V6

435
00:14:12,020 --> 00:14:13,260
自动生成就可以

436
00:14:13,260 --> 00:14:16,580
尽管说它对于复杂场景的表现不如Synthense

437
00:14:16,580 --> 00:14:19,080
但是价格是Synthense的至少七分之一

438
00:14:19,080 --> 00:14:20,880
打折的时候能到十分之一

439
00:14:20,880 --> 00:14:22,100
而且精度我试过

440
00:14:22,100 --> 00:14:23,720
很多运镜也能胜任

441
00:14:23,720 --> 00:14:25,260
尽管有时候会有小问题

442
00:14:25,260 --> 00:14:27,160
但大部分时候完全够用

443
00:14:27,160 --> 00:14:28,760
比如说一个视频300个分镜

444
00:14:28,760 --> 00:14:30,300
我们其中60个镜头

445
00:14:30,300 --> 00:14:31,760
可以用Sindance或者H3

446
00:14:31,760 --> 00:14:33,480
已经有五分之一的比例了

447
00:14:33,480 --> 00:14:35,660
而其余的我们全部都用PixWall

448
00:14:35,660 --> 00:14:38,240
每五个镜头里面就有一个高级运镜

449
00:14:38,240 --> 00:14:39,780
其实已经非常足够了

450
00:14:39,780 --> 00:14:40,940
你在这个画布上

451
00:14:40,940 --> 00:14:42,660
任何一个生成出来的东西

452
00:14:42,660 --> 00:14:44,100
你可以选成标注

453
00:14:44,100 --> 00:14:44,700
选用

454
00:14:44,700 --> 00:14:46,000
废弃或者带复核

455
00:14:46,000 --> 00:14:47,060
可以自动分区

456
00:14:47,060 --> 00:14:47,860
项目一大

457
00:14:47,860 --> 00:14:48,540
素材

458
00:14:48,540 --> 00:14:49,020
脚本

459
00:14:49,020 --> 00:14:49,580
角色

460
00:14:49,580 --> 00:14:50,160
分镜

461
00:14:50,160 --> 00:14:51,280
视频镜头

462
00:14:51,280 --> 00:14:52,140
最终选片

463
00:14:52,140 --> 00:14:53,940
都会各自待在各自的区域

464
00:14:53,940 --> 00:14:54,900
不会堆成一坨

465
00:14:54,900 --> 00:14:56,320
这个听起来是小事

466
00:14:56,320 --> 00:14:58,100
但是做到第一百个分镜的时候

467
00:14:58,100 --> 00:14:59,440
你就知道它有多重要

468
00:14:59,440 --> 00:15:01,060
每一个生成出来的节点

469
00:15:01,060 --> 00:15:02,840
也都记着它当时用的提示词

470
00:15:02,840 --> 00:15:03,680
哪个模型

471
00:15:03,680 --> 00:15:04,400
什么参数

472
00:15:04,400 --> 00:15:05,940
也就是说在三个月之后

473
00:15:05,940 --> 00:15:07,720
你想要去复现某一条

474
00:15:07,720 --> 00:15:08,860
效果特别好的画面

475
00:15:08,860 --> 00:15:10,960
你都查得到当时是怎么跑出来的

476
00:15:10,960 --> 00:15:12,780
我用完以后最大的感受就是

477
00:15:12,780 --> 00:15:14,180
它把管理这件事

478
00:15:14,180 --> 00:15:15,760
从你的脑子里面抽走了

479
00:15:15,760 --> 00:15:18,380
你可以把注意力放回内容本身

480
00:15:18,380 --> 00:15:19,440
可能又会问了

481
00:15:19,440 --> 00:15:23,160
那我可以用无相话簿熟悉和学习做AI视频

482
00:15:23,160 --> 00:15:26,280
但是你之前所说的自动化工作流怎么样做呢

483
00:15:26,280 --> 00:15:29,000
这个呢是属于我们做AI视频的高端玩法

484
00:15:29,000 --> 00:15:31,340
你可以直接在电脑的终端里面完成

485
00:15:31,340 --> 00:15:32,860
也并不需要其他的工具

486
00:15:32,860 --> 00:15:34,620
Pixworks有CLI的接口

487
00:15:34,620 --> 00:15:36,080
只要你输入这行指令

488
00:15:36,080 --> 00:15:38,580
你就可以直接调用Pixworks到你的终端

489
00:15:38,580 --> 00:15:39,740
排队集体生成

490
00:15:39,740 --> 00:15:40,440
打个比方

491
00:15:40,440 --> 00:15:42,660
刚刚我们用的网页图形界面

492
00:15:42,660 --> 00:15:45,740
就像你在餐厅里面点菜是一道一道的点

493
00:15:45,740 --> 00:15:46,920
等上了一道之后

494
00:15:46,920 --> 00:15:48,200
你才开始点下一道

495
00:15:48,200 --> 00:15:52,460
而CLI就像你把一张写好200道菜的单子递进厨房

496
00:15:52,460 --> 00:15:54,860
前者你需要说200次等200次

497
00:15:54,860 --> 00:15:56,520
后者你只需要递一次

498
00:15:56,520 --> 00:15:57,980
然后他一次性给你上菜

499
00:15:57,980 --> 00:15:59,660
在这里登录也是一个命令

500
00:16:00,000 --> 00:16:06,000
在這裡登錄也是一個命令,瀏覽器談一個授權頁,你點確認,不用手動去複製什麼API密钥,用的就是你自己的賬號。

501
00:16:06,000 --> 00:16:15,000
額度,會員權益,素材庫全部都是通的。我們在終端輸入一條生圖指令,這個圖就會開始生成,一鍵調用,比網頁版更加方便。

502
00:16:15,000 --> 00:16:15,840
证程完了之后

503
00:16:15,840 --> 00:16:16,840
它会给你一个网址

504
00:16:16,840 --> 00:16:18,240
你可以自己去下载

505
00:16:18,240 --> 00:16:19,240
你已经搞好了

506
00:16:19,240 --> 00:16:21,640
你也可以让你的AI自动的下载下来

507
00:16:21,640 --> 00:16:22,180
刚刚呢

508
00:16:22,180 --> 00:16:23,420
我们展示是在终端

509
00:16:23,420 --> 00:16:26,960
但是这个Cli是可以接到你喜欢的任何一个AI上的

510
00:16:26,960 --> 00:16:28,740
官方明确支持Cloud Code

511
00:16:28,740 --> 00:16:29,420
Codex

512
00:16:29,420 --> 00:16:30,140
OpenCloud

513
00:16:30,140 --> 00:16:31,040
Permas这些

514
00:16:31,040 --> 00:16:33,720
就是直接跟你的AI说要安装就可以

515
00:16:33,720 --> 00:16:35,200
如果你嫌配置麻烦

516
00:16:35,200 --> 00:16:36,700
还有更省事的方法

517
00:16:36,700 --> 00:16:38,000
官方把安装步骤

518
00:16:38,000 --> 00:16:38,980
所有的命令

519
00:16:38,980 --> 00:16:39,620
参数

520
00:16:39,620 --> 00:16:40,360
模型

521
00:16:40,360 --> 00:16:40,940
限制

522
00:16:40,940 --> 00:16:42,820
写成了一个纯文本的文件

523
00:16:42,820 --> 00:16:45,120
你可以直接让你的AI去读这个地址

524
00:16:45,120 --> 00:16:46,900
它读完之后自己就会配了

525
00:16:46,900 --> 00:16:48,320
你一行文档都不用看

526
00:16:48,320 --> 00:16:50,700
你可以让它生成提示词studyboard

527
00:16:50,700 --> 00:16:53,480
然后再让它一口气生成所有的图片

528
00:16:53,480 --> 00:16:54,960
按照文件批量跑

529
00:16:54,960 --> 00:16:56,380
你可以直接去干别的

530
00:16:56,380 --> 00:16:57,640
任务在后台排队

531
00:16:57,640 --> 00:17:00,320
全部渲染好之后会落进你的素材库

532
00:17:00,320 --> 00:17:02,340
就算你把中章关了也不会丢

533
00:17:02,340 --> 00:17:04,080
失败的任务它会记状态

534
00:17:04,080 --> 00:17:05,320
你可以单独查

535
00:17:05,320 --> 00:17:06,140
单独重跑

536
00:17:06,140 --> 00:17:06,860
单独审核

537
00:17:06,860 --> 00:17:08,080
不需要整批重来

538
00:17:08,080 --> 00:17:10,020
做200多个分间的长视频

539
00:17:10,020 --> 00:17:11,140
这个是刚需

540
00:17:11,140 --> 00:17:12,440
不是锦上添花

541
00:17:12,440 --> 00:17:13,420
素材都在了

542
00:17:13,420 --> 00:17:15,560
剩下的就把它直接拼下来就行

543
00:17:15,560 --> 00:17:16,760
也就是组装一下

544
00:17:16,760 --> 00:17:17,580
就可以直出

545
00:17:17,580 --> 00:17:18,420
第二个赛道

546
00:17:18,420 --> 00:17:19,200
也是在今年

547
00:17:19,200 --> 00:17:21,180
在YouTube上特别火的一类

548
00:17:21,180 --> 00:17:23,100
就是吉普利风格的怀旧动画

549
00:17:23,100 --> 00:17:24,480
这类账号轻轻松松

550
00:17:24,480 --> 00:17:25,920
就能斩获大量的流量

551
00:17:25,920 --> 00:17:28,040
你看这个频道发布了50多个视频

552
00:17:28,040 --> 00:17:30,240
就斩获了900多万次的观看

553
00:17:30,240 --> 00:17:31,700
那种形式非常简单

554
00:17:31,700 --> 00:17:34,340
用AI做出来那种温暖治愈的画风

555
00:17:34,340 --> 00:17:36,460
展示某个年代某个地方的日常

556
00:17:36,460 --> 00:17:37,680
90年代的乡村

557
00:17:37,680 --> 00:17:38,320
雨季

558
00:17:38,320 --> 00:17:40,140
一家人在灶台前面做饭

559
00:17:40,140 --> 00:17:41,380
火车穿过田野

560
00:17:41,380 --> 00:17:42,320
没有旁白

561
00:17:42,320 --> 00:17:43,000
对话

562
00:17:43,000 --> 00:17:44,820
就是画面和环境音

563
00:17:44,820 --> 00:17:47,300
这一批账号有一个非常明显的共性

564
00:17:47,300 --> 00:17:48,380
你去看就知道了

565
00:17:48,380 --> 00:17:49,700
全部都是长视频

566
00:17:49,700 --> 00:17:51,080
五分钟到十几分钟

567
00:17:51,080 --> 00:17:52,020
而不是短视频

568
00:17:52,020 --> 00:17:54,380
这一点是这个赛道能够赚钱的核心

569
00:17:54,380 --> 00:17:56,560
因为这类内容的观众不是刷的

570
00:17:56,560 --> 00:17:57,380
是放着的

571
00:17:57,380 --> 00:17:59,920
很多人在睡前当做白噪音来听

572
00:17:59,920 --> 00:18:01,420
停留时间非常长

573
00:18:01,420 --> 00:18:02,580
这样广告位就多

574
00:18:02,580 --> 00:18:03,600
同样是不露脸

575
00:18:03,600 --> 00:18:05,300
做长视频的怀旧动画号

576
00:18:05,300 --> 00:18:07,840
收益能比很多的短视频号高一个量级

577
00:18:07,840 --> 00:18:08,940
但是问题也来了

578
00:18:08,940 --> 00:18:10,240
一条十分钟的动画

579
00:18:10,240 --> 00:18:11,700
大概需要两百个分镜

580
00:18:11,700 --> 00:18:12,500
200张图

581
00:18:12,500 --> 00:18:13,420
200段视频

582
00:18:13,420 --> 00:18:16,300
还有保证从第一个分镜到第200个分镜

583
00:18:16,300 --> 00:18:17,420
主角的脸不变

584
00:18:17,420 --> 00:18:18,300
衣服不变

585
00:18:18,300 --> 00:18:19,020
画风不变

586
00:18:19,020 --> 00:18:21,380
这个工作量才是真正劝退人的地方

587
00:18:21,380 --> 00:18:23,920
那在这里呢就要提到顺劲非常的重要

588
00:18:23,920 --> 00:18:25,940
很多人一上来就升分镜图

589
00:18:25,940 --> 00:18:27,340
到了第三张图的时候

590
00:18:27,340 --> 00:18:29,240
主角已经换了第三张脸了

591
00:18:29,240 --> 00:18:31,340
正确的顺序是先把角色钉死

592
00:18:31,340 --> 00:18:33,800
再让所有的分镜都参考这一套角色

593
00:18:33,800 --> 00:18:35,880
角色的数量控制再4到6个

594
00:18:35,880 --> 00:18:37,040
少的话故事单调

595
00:18:37,040 --> 00:18:39,460
多了后面维持一致性会非常痛苦

596
00:18:39,460 --> 00:18:40,940
无变的特征是核心

597
00:18:40,940 --> 00:18:42,760
AI维持角色一致性

598
00:18:42,760 --> 00:18:44,600
靠的是抓住一个显著的

599
00:18:44,600 --> 00:18:45,620
能画出来的锚点

600
00:18:45,620 --> 00:18:46,860
你写温柔的母亲

601
00:18:46,860 --> 00:18:47,800
她抓不住

602
00:18:47,800 --> 00:18:48,740
但是你可以写

603
00:18:48,740 --> 00:18:51,260
左肩上永远搭着一条褪色的蓝毛巾

604
00:18:51,260 --> 00:18:54,000
这样她每张图就会把这条毛巾给画上去

605
00:18:54,000 --> 00:18:56,020
观众也会靠这条毛巾来认人

606
00:18:56,020 --> 00:18:58,440
然后我们在这里选择添加图片节点

607
00:18:58,440 --> 00:18:59,860
可以看到这里的提示词

608
00:18:59,860 --> 00:19:01,260
已经全部都预设好了

609
00:19:01,260 --> 00:19:02,640
因为是连在一起的

610
00:19:02,640 --> 00:19:04,100
直接让它生成就可以

611
00:19:04,100 --> 00:19:05,600
角色图生成好之后

612
00:19:05,600 --> 00:19:06,560
我们不下载

613
00:19:06,560 --> 00:19:07,760
直接留在画布上

614
00:19:07,760 --> 00:19:08,880
这个是无线画布

615
00:19:08,880 --> 00:19:11,300
跟摄政完就下载这个模式最大的区别

616
00:19:11,300 --> 00:19:14,380
这五张角色图在画布上就是五个参考节点

617
00:19:14,380 --> 00:19:15,780
你后面每做一个分镜

618
00:19:15,780 --> 00:19:18,680
只要把线从角色节点拉到分镜节点

619
00:19:18,680 --> 00:19:20,060
形象就自动带进去了

620
00:19:20,060 --> 00:19:22,400
不需要每次重新上传参考图

621
00:19:22,400 --> 00:19:24,520
也不需要每次在提示词里面

622
00:19:24,520 --> 00:19:26,460
重新描述一遍这个人长什么样

623
00:19:26,460 --> 00:19:28,200
配合前面说的项目记忆

624
00:19:28,200 --> 00:19:29,840
你做到第百个分镜的时候

625
00:19:29,840 --> 00:19:31,760
画风和角色依然是稳的

626
00:19:31,760 --> 00:19:34,740
你可以直接让AI帮你把分镜表给生成出来

627
00:19:34,740 --> 00:19:36,560
然后让它一口气的帮你生图

628
00:19:36,560 --> 00:19:39,320
不过不要让AI一次性生成超过20条

629
00:19:39,320 --> 00:19:41,540
不然它到了后面会明显开始敷衍

630
00:19:41,540 --> 00:19:42,600
开始写重复的

631
00:19:42,600 --> 00:19:43,840
没有信息量的场景

632
00:19:43,840 --> 00:19:44,880
拿到分信之后

633
00:19:44,880 --> 00:19:46,860
你可以直接用无线化布生成

634
00:19:46,860 --> 00:19:48,720
也可以用刚才降约CLI

635
00:19:48,720 --> 00:19:49,820
批量跑就行了

636
00:19:49,820 --> 00:19:52,200
200张图和200段视频一次提交

637
00:19:52,200 --> 00:19:54,760
回来的时候东西已经在文件夹里了

638
00:19:54,760 --> 00:19:56,420
我们再来说到第三个赛道

639
00:19:56,420 --> 00:19:58,060
如果你是刚刚起步的小白

640
00:19:58,060 --> 00:19:59,020
我尤其推荐

641
00:19:59,020 --> 00:20:00,520
大家可以看这类频道

642
00:20:00,520 --> 00:20:02,320
讲的是远古人类的故事

643
00:20:02,320 --> 00:20:03,700
原始人怎么熬过冬天

644
00:20:03,700 --> 00:20:05,380
没有火之前他们怎么活

645
00:20:05,380 --> 00:20:07,220
猛兽时代怎么生存下来

646
00:20:07,220 --> 00:20:10,100
90个视频已经获得了400多万次的观看

647
00:20:10,100 --> 00:20:12,360
变现至少在15000美金里上

648
00:20:12,360 --> 00:20:14,800
你现在看一下它的画面就会发现一件事

649
00:20:14,800 --> 00:20:15,840
它甚至没有动画

650
00:20:15,840 --> 00:20:18,460
主体就是一张一张AI生成的图片在切

651
00:20:18,460 --> 00:20:19,280
非常旁白

652
00:20:19,280 --> 00:20:21,080
这个意味着你做这类视频

653
00:20:21,080 --> 00:20:23,980
你不需要去处理任何跟视频生成有关的麻烦事

654
00:20:23,980 --> 00:20:25,020
你不用管运镜

655
00:20:25,020 --> 00:20:27,560
不用管角色在视频里面会不会变形

656
00:20:27,560 --> 00:20:28,580
物理是不是合理

657
00:20:28,580 --> 00:20:30,840
你只需要有一批质量足够好的图

658
00:20:30,840 --> 00:20:32,940
再加上一段旁白就能出片

659
00:20:32,940 --> 00:20:34,140
根据这个运营思路

660
00:20:34,140 --> 00:20:36,040
也有很多其他的方向和赛道

661
00:20:36,040 --> 00:20:38,280
比如说这个频道是做历史动画的

662
00:20:38,280 --> 00:20:40,520
9个视频就拿到了几十万的播放量

663
00:20:40,520 --> 00:20:42,640
视频也是用图片的形式来切换

664
00:20:42,640 --> 00:20:44,500
其实就是把一些历史小知识

665
00:20:44,500 --> 00:20:46,400
用比较诙谐的方式讲出来

666
00:20:46,400 --> 00:20:47,960
这才饱和度就非常高

667
00:20:47,960 --> 00:20:48,520
很漂亮

668
00:20:48,520 --> 00:20:50,360
但是该有的历史元素也都有

669
00:20:50,360 --> 00:20:51,580
你要说门槛的话

670
00:20:51,580 --> 00:20:53,800
比之前做视频的都低了一大截

671
00:20:53,800 --> 00:20:55,820
但是门槛低不代表随便做

672
00:20:55,820 --> 00:20:57,780
那这类视频呢有两个喜学

673
00:20:57,780 --> 00:20:59,640
首先是图片不能出错

674
00:20:59,640 --> 00:21:01,220
在动态视频里面

675
00:21:01,220 --> 00:21:02,360
画面一直都在动

676
00:21:02,360 --> 00:21:04,060
AI就算有犯的小错误

677
00:21:04,060 --> 00:21:05,300
观众也来不及看

678
00:21:05,300 --> 00:21:06,620
但是性态图不一样

679
00:21:06,620 --> 00:21:08,620
一张图就算只停在那两秒

680
00:21:08,620 --> 00:21:10,660
观众也有充足的时间盯着看

681
00:21:10,660 --> 00:21:11,680
手指有几根

682
00:21:11,680 --> 00:21:12,680
眼睛对不对称

683
00:21:12,680 --> 00:21:13,620
透视有没有崩

684
00:21:13,620 --> 00:21:14,700
全部都藏不住

685
00:21:14,700 --> 00:21:16,060
所以这个赛道的图

686
00:21:16,060 --> 00:21:18,100
需要用不同的思路来生成

687
00:21:18,100 --> 00:21:19,220
我总结了三条

688
00:21:19,220 --> 00:21:21,160
你可以直接加到提示词里面

689
00:21:21,160 --> 00:21:22,300
让AI作为规范

690
00:21:22,300 --> 00:21:23,900
在生提示词的时候

691
00:21:23,900 --> 00:21:25,600
主动加遮挡和前颈

692
00:21:25,600 --> 00:21:27,440
让人物的手里面拿着东西

693
00:21:27,440 --> 00:21:28,860
身上披着兽皮

694
00:21:28,860 --> 00:21:31,500
前面有草丛或者岩石挡一部分

695
00:21:31,500 --> 00:21:32,320
遮挡越多

696
00:21:32,320 --> 00:21:34,400
AI需要正确渲染的面积越小

697
00:21:34,400 --> 00:21:36,020
出色的概率就越低

698
00:21:36,020 --> 00:21:37,380
还有就是统一光源

699
00:21:37,380 --> 00:21:38,820
整条视频的所有图片

700
00:21:38,820 --> 00:21:40,280
光线都一定要一样

701
00:21:40,280 --> 00:21:42,660
这个是让200张分别生成出来的图

702
00:21:42,660 --> 00:21:45,240
看起来像同一条片子的最简单方法

703
00:21:45,240 --> 00:21:46,440
比调色还要有效

704
00:21:46,440 --> 00:21:48,320
我把这三条写进了提示词

705
00:21:48,320 --> 00:21:49,660
大家可以直接照抄

706
00:21:49,660 --> 00:21:52,460
这段提示词跟前面两个赛道最大的区别

707
00:21:52,460 --> 00:21:54,300
是它不是在描述画什么

708
00:21:54,300 --> 00:21:56,220
而是在规定怎么样构图

709
00:21:56,220 --> 00:21:57,740
才能避开AI的弱点

710
00:21:57,740 --> 00:22:00,400
这个赛道不要指望AI画的越来越好

711
00:22:00,400 --> 00:22:03,560
而是要主动把画面设计成它不容易出错的样子

712
00:22:03,560 --> 00:22:05,080
还有就是切图的节奏

713
00:22:05,080 --> 00:22:07,880
自态图片在屏幕上停超过6秒钟

714
00:22:07,880 --> 00:22:09,280
观众就会开始走神

715
00:22:09,280 --> 00:22:11,100
手指就开始往下滑了

716
00:22:11,100 --> 00:22:12,300
这个不是我猜的

717
00:22:12,300 --> 00:22:13,720
你自己感受一下就知道

718
00:22:13,720 --> 00:22:15,440
移动不动的图看6秒钟

719
00:22:15,440 --> 00:22:17,260
你的注意力早就已经飘了

720
00:22:17,260 --> 00:22:20,320
所以说我会把一张图控制在1.5到5秒之间

721
00:22:20,320 --> 00:22:22,160
也就是说一条10分钟的视频

722
00:22:22,160 --> 00:22:24,080
你需要250到300张图

723
00:22:24,080 --> 00:22:27,160
这个数量手动一张张生成是不可能的

724
00:22:27,160 --> 00:22:28,180
必须要批量跑

725
00:22:28,180 --> 00:22:29,760
所以说一定要一口气

726
00:22:29,760 --> 00:22:31,400
就能够调用N多模型

727
00:22:31,400 --> 00:22:33,180
切来切去的根本不现实

728
00:22:33,180 --> 00:22:36,440
在这里尤其适合你用的是Pixels CLI

729
00:22:36,440 --> 00:22:37,460
可以一键生成

730
00:22:37,460 --> 00:22:38,680
包括配音旁白

731
00:22:38,680 --> 00:22:41,780
直接用Pixels Create Voice生成就行

732
00:22:41,780 --> 00:22:43,400
跟图片在同一个账号

733
00:22:43,400 --> 00:22:44,420
同一套额度里

734
00:22:44,420 --> 00:22:46,400
不用再单开一个配音网站

735
00:22:46,400 --> 00:22:47,600
如果你零基础

736
00:22:47,600 --> 00:22:49,400
可以直接从第三个开始

737
00:22:49,400 --> 00:22:51,640
也就是远古人故事这种图片流

738
00:22:51,640 --> 00:22:54,060
因为它不涉及视频生成的任何麻烦

739
00:22:54,060 --> 00:22:55,940
你只需要去跑图和剪辑

740
00:22:55,940 --> 00:22:57,340
一天就能出第一条

741
00:22:57,340 --> 00:22:58,900
基本上只要你图出了

742
00:22:58,900 --> 00:22:59,900
视频也就有了

743
00:22:59,900 --> 00:23:00,820
加两个方向

744
00:23:00,820 --> 00:23:02,800
如果你感兴趣也都可以尝试

745
00:23:02,800 --> 00:23:03,700
说实话哈

746
00:23:03,700 --> 00:23:06,140
毕竟这些频道全部都是不露脸的

747
00:23:06,140 --> 00:23:09,400
现在你也可以用Pixels一键去调用所有的工具

748
00:23:09,400 --> 00:23:11,420
那其实当你做完一个账号之后

749
00:23:11,420 --> 00:23:12,280
你再做一个

750
00:23:12,280 --> 00:23:14,940
甚至同时做两三个账号都是有可能的

751
00:23:14,940 --> 00:23:16,760
我们来回头看这整一套流程

752
00:23:16,760 --> 00:23:20,560
真正的变化其实并不在于某一个AI单独变强了

753
00:23:20,560 --> 00:23:23,240
而是在于这些能力都可以被收到同一个地方

754
00:23:23,240 --> 00:23:25,480
以前我做一条视频要开五个工具

755
00:23:25,480 --> 00:23:27,300
现在从一个想法到提示词

756
00:23:27,300 --> 00:23:28,020
到生成

757
00:23:28,020 --> 00:23:28,740
到成片

758
00:23:28,740 --> 00:23:30,100
都是一条线走完的

759
00:23:30,100 --> 00:23:32,900
这个差别听起来只是省了几次复制粘贴

760
00:23:32,900 --> 00:23:35,600
但是对于一个想要长期做频道的人来讲

761
00:23:35,600 --> 00:23:38,420
它省下来的是最宝贵的那一部分的注意力

762
00:23:38,420 --> 00:23:40,200
你不需要把利息给花在

763
00:23:40,200 --> 00:23:41,700
那我到底做到哪了

764
00:23:41,700 --> 00:23:44,360
而是花在我下一条究竟应该做什么上

765
00:23:44,360 --> 00:23:45,780
如果你想要去试一试

766
00:23:45,780 --> 00:23:48,140
BigSource的链接在描述栏第一条

767
00:23:48,140 --> 00:23:49,920
网页端和命令行都可以用

768
00:23:49,920 --> 00:23:52,280
你可以直接在描述栏注册和下载

769
00:23:52,280 --> 00:23:53,200
有专属折扣

770
00:23:53,200 --> 00:23:55,180
也欢迎在评论区里面告诉我

771
00:23:55,180 --> 00:23:56,320
这三个赛道里面

772
00:23:56,320 --> 00:23:57,800
今天你最想要做哪一个

773
00:23:57,800 --> 00:23:59,000
对哪个最感兴趣

774
00:23:59,000 --> 00:23:59,860
或者说你有什么

775
00:23:59,860 --> 00:24:01,420
更感兴趣的其他方向

776
00:24:01,420 --> 00:24:02,560
那如果说人多的话

777
00:24:02,560 --> 00:24:04,200
我后面专门出视频来讲解

778
00:24:04,200 --> 00:24:04,720
我是安然

779
00:24:04,720 --> 00:24:05,820
也欢迎大家订阅我

780
00:24:05,820 --> 00:24:07,100
按下的旁边的小微单

781
00:24:07,100 --> 00:24:07,560
点赞

782
00:24:07,560 --> 00:24:08,580
这样你就不会错过

783
00:24:08,580 --> 00:24:10,280
我们后期任何有价值的内容啦

784
00:24:10,280 --> 00:24:11,060
那我们下期见

785
00:24:11,060 --> 00:24:11,380
拜拜