@lincyaw/skills 1.0.0-rc.2 → 1.0.0-rc.21
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +1 -1
- package/package.json +2 -3
- package/skills/index.json +9 -18
- package/skills/presmium-animate/SKILL.md +4 -2
- package/skills/presmium-animate/gallery/algorithms/lecture.json +171 -115
- package/skills/presmium-animate/gallery/anime/scenes/kmeans-seeded.ts +2 -1
- package/skills/presmium-animate/gallery/craft/lecture.json +109 -57
- package/skills/presmium-animate/gallery/data/lecture.json +711 -175
- package/skills/presmium-animate/gallery/emphasis/lecture.json +321 -79
- package/skills/presmium-animate/gallery/films/lecture.json +36 -10
- package/skills/presmium-animate/gallery/lengths/lecture.json +23 -88
- package/skills/presmium-animate/gallery/math-ml/lecture.json +238 -159
- package/skills/presmium-animate/gallery/paths-3d/lecture.json +74 -64
- package/skills/presmium-animate/gallery/product/lecture.json +27 -3
- package/skills/presmium-animate/gallery/story/lecture.json +23 -90
- package/skills/presmium-animate/gallery/structure/lecture.json +104 -65
- package/skills/presmium-animate/gallery/text/lecture.json +115 -63
- package/skills/presmium-animate/gallery/timeline/lecture.json +106 -187
- package/skills/presmium-animate/gallery/timeline/scenes/bubble.ts +1 -1
- package/skills/presmium-animate/gallery/transitions/lecture.json +103 -52
- package/skills/presmium-animate/references/lengths.md +1 -1
- package/skills/presmium-animate/references/libraries.md +12 -6
- package/skills/presmium-animate/references/procedural.md +1 -0
- package/skills/presmium-animate/references/story.md +21 -7
- package/skills/presmium-animate/references/team.md +176 -0
- package/skills/presmium-author/SKILL.md +53 -19
- package/skills/presmium-author/references/assistance.md +3 -1
- package/skills/{presmium-present → presmium-author}/references/classroom.md +2 -2
- package/skills/presmium-author/references/document.md +1 -1
- package/skills/{presmium-present → presmium-author}/references/lecture.md +3 -1
- package/skills/presmium-author/references/publishing.md +161 -13
- package/skills/presmium-author/references/writing.md +170 -7
- package/src/cli.ts +4 -4
- package/src/install.ts +42 -5
- package/src/messages.ts +5 -5
- package/src/resources.ts +5 -4
- package/skills/presmium-present/SKILL.md +0 -35
- package/skills/presmium-present/references/delivery.md +0 -122
|
@@ -99,7 +99,7 @@
|
|
|
99
99
|
"children": [
|
|
100
100
|
{
|
|
101
101
|
"type": "text",
|
|
102
|
-
"text": "
|
|
102
|
+
"text": "一段带讲解的动画,动手写代码之前先写工作单:主旨、中心画面、叙事形状和分镜表。这份文稿用一段 flow matching 动画示范这张工作单,写法见 "
|
|
103
103
|
},
|
|
104
104
|
{
|
|
105
105
|
"type": "text",
|
|
@@ -112,7 +112,7 @@
|
|
|
112
112
|
},
|
|
113
113
|
{
|
|
114
114
|
"type": "text",
|
|
115
|
-
"text": "
|
|
115
|
+
"text": "。分镜表的每一行是动画的一个状态,每个状态停住的那一帧单独看也能读懂。"
|
|
116
116
|
}
|
|
117
117
|
]
|
|
118
118
|
},
|
|
@@ -122,7 +122,7 @@
|
|
|
122
122
|
"children": [
|
|
123
123
|
{
|
|
124
124
|
"type": "text",
|
|
125
|
-
"text": "
|
|
125
|
+
"text": "叙事参考小红书上的讲解视频《flow matching:将生图路径,拉直!》:噪声到图的路径这一个画面贯穿始终,每个新想法都是它的一次变化。"
|
|
126
126
|
}
|
|
127
127
|
]
|
|
128
128
|
}
|
|
@@ -138,16 +138,7 @@
|
|
|
138
138
|
"children": [
|
|
139
139
|
{
|
|
140
140
|
"type": "text",
|
|
141
|
-
"text": "
|
|
142
|
-
"marks": [
|
|
143
|
-
{
|
|
144
|
-
"type": "bold"
|
|
145
|
-
}
|
|
146
|
-
]
|
|
147
|
-
},
|
|
148
|
-
{
|
|
149
|
-
"type": "text",
|
|
150
|
-
"text": ":看完后,观众明白 flow matching 把生成的路径拉直,所以一两步就能生成,因为他们看到同一批粒子沿着弯路要走几十步,拉直之后一步就落进标志。"
|
|
141
|
+
"text": "主旨:看完后,观众明白 flow matching 把生成的路径拉直,所以一两步就能生成;依据是同一批粒子沿弯路要走几十步,拉直之后一步就落进标志。"
|
|
151
142
|
}
|
|
152
143
|
]
|
|
153
144
|
},
|
|
@@ -157,16 +148,7 @@
|
|
|
157
148
|
"children": [
|
|
158
149
|
{
|
|
159
150
|
"type": "text",
|
|
160
|
-
"text": "ABT"
|
|
161
|
-
"marks": [
|
|
162
|
-
{
|
|
163
|
-
"type": "bold"
|
|
164
|
-
}
|
|
165
|
-
]
|
|
166
|
-
},
|
|
167
|
-
{
|
|
168
|
-
"type": "text",
|
|
169
|
-
"text": ":生成就是让噪声沿着路径走到图,而且网络只需要学每一处的速度;但是路是弯的,步数少就走不准;所以用直线训练,再用 Reflow 把路拉直。"
|
|
151
|
+
"text": "ABT(And, But, Therefore):生成是让噪声沿路径走到图,网络只需学每一处的速度;但是路是弯的,步数少就走不准;所以先用直线训练,再用 Reflow 把路拉直。"
|
|
170
152
|
}
|
|
171
153
|
]
|
|
172
154
|
},
|
|
@@ -176,29 +158,7 @@
|
|
|
176
158
|
"children": [
|
|
177
159
|
{
|
|
178
160
|
"type": "text",
|
|
179
|
-
"text": "
|
|
180
|
-
"marks": [
|
|
181
|
-
{
|
|
182
|
-
"type": "bold"
|
|
183
|
-
}
|
|
184
|
-
]
|
|
185
|
-
},
|
|
186
|
-
{
|
|
187
|
-
"type": "text",
|
|
188
|
-
"text": ":三维的时空,时间是第三根轴:噪声平面(t = 0)和目标平面(t = 1)沿它相对,每个粒子的路径是两平面之间的一条曲线,下面一条时间轴。"
|
|
189
|
-
},
|
|
190
|
-
{
|
|
191
|
-
"type": "text",
|
|
192
|
-
"text": "最有说服力的一帧",
|
|
193
|
-
"marks": [
|
|
194
|
-
{
|
|
195
|
-
"type": "bold"
|
|
196
|
-
}
|
|
197
|
-
]
|
|
198
|
-
},
|
|
199
|
-
{
|
|
200
|
-
"type": "text",
|
|
201
|
-
"text": ":最后两块面板,同样走 1 步,一团和一个标志。"
|
|
161
|
+
"text": "中心画面:三维时空,时间是第三根轴。噪声平面(t = 0)和目标平面(t = 1)沿时间轴相对,每个粒子的路径是两平面之间的一条曲线,画面下方是一条时间轴。最有说服力的一帧是最后两块面板:同样走 1 步,一边挤成一团,一边画出标志。"
|
|
202
162
|
}
|
|
203
163
|
]
|
|
204
164
|
},
|
|
@@ -208,16 +168,7 @@
|
|
|
208
168
|
"children": [
|
|
209
169
|
{
|
|
210
170
|
"type": "text",
|
|
211
|
-
"text": "
|
|
212
|
-
"marks": [
|
|
213
|
-
{
|
|
214
|
-
"type": "bold"
|
|
215
|
-
}
|
|
216
|
-
]
|
|
217
|
-
},
|
|
218
|
-
{
|
|
219
|
-
"type": "text",
|
|
220
|
-
"text": ":两幕。问题(弯路要很多步)→ 机制(学速度、按步积分)→ 答案(直线训练)→ 遗留问题(直线交叉,平均方向让路又弯)→ 第二个答案(Reflow)。"
|
|
171
|
+
"text": "叙事形状分两幕:问题(弯路要走很多步)、机制(学速度、按步积分)、答案(直线训练)、遗留问题(直线交叉,取平均的方向让路又弯)、第二个答案(Reflow)。"
|
|
221
172
|
}
|
|
222
173
|
]
|
|
223
174
|
},
|
|
@@ -227,16 +178,7 @@
|
|
|
227
178
|
"children": [
|
|
228
179
|
{
|
|
229
180
|
"type": "text",
|
|
230
|
-
"text": "
|
|
231
|
-
"marks": [
|
|
232
|
-
{
|
|
233
|
-
"type": "bold"
|
|
234
|
-
}
|
|
235
|
-
]
|
|
236
|
-
},
|
|
237
|
-
{
|
|
238
|
-
"type": "text",
|
|
239
|
-
"text": ":红色是真实(精确路径、训练目标),强调色是网络的箭头,黄色是采样器实际走的,绿色是拉直后的模型。"
|
|
181
|
+
"text": "每种颜色只有一种含义:红色是真实值(精确路径、训练目标),强调色是网络预测的箭头,黄色是采样器实际走的路径,绿色是拉直后的模型。"
|
|
240
182
|
}
|
|
241
183
|
]
|
|
242
184
|
},
|
|
@@ -246,16 +188,7 @@
|
|
|
246
188
|
"children": [
|
|
247
189
|
{
|
|
248
190
|
"type": "text",
|
|
249
|
-
"text": "
|
|
250
|
-
"marks": [
|
|
251
|
-
{
|
|
252
|
-
"type": "bold"
|
|
253
|
-
}
|
|
254
|
-
]
|
|
255
|
-
},
|
|
256
|
-
{
|
|
257
|
-
"type": "text",
|
|
258
|
-
"text": ":"
|
|
191
|
+
"text": "数字来自 "
|
|
259
192
|
},
|
|
260
193
|
{
|
|
261
194
|
"type": "text",
|
|
@@ -353,7 +286,7 @@
|
|
|
353
286
|
"children": [
|
|
354
287
|
{
|
|
355
288
|
"type": "text",
|
|
356
|
-
"text": "
|
|
289
|
+
"text": "生成一张图,就是把一团随机噪声推成图。每个噪声点沿着自己的路,走到图里属于它的位置。"
|
|
357
290
|
}
|
|
358
291
|
]
|
|
359
292
|
},
|
|
@@ -399,7 +332,7 @@
|
|
|
399
332
|
"children": [
|
|
400
333
|
{
|
|
401
334
|
"type": "text",
|
|
402
|
-
"text": "
|
|
335
|
+
"text": "网络不学整条路,只学每个位置、每个时刻该往哪走,也就是速度。速度就是路径的切线。"
|
|
403
336
|
}
|
|
404
337
|
]
|
|
405
338
|
},
|
|
@@ -445,7 +378,7 @@
|
|
|
445
378
|
"children": [
|
|
446
379
|
{
|
|
447
380
|
"type": "text",
|
|
448
|
-
"text": "
|
|
381
|
+
"text": "生成时,把时间切成几段,每段沿着速度走一步。可路是弯的,每一步都会偏出去一点。四步偏得明显,三十二步才贴住。"
|
|
449
382
|
}
|
|
450
383
|
]
|
|
451
384
|
},
|
|
@@ -537,7 +470,7 @@
|
|
|
537
470
|
"children": [
|
|
538
471
|
{
|
|
539
472
|
"type": "text",
|
|
540
|
-
"text": "Flow matching
|
|
473
|
+
"text": "Flow matching 的训练很简单。随便配一个噪声点和一个图上的点,连一条直线。在线上随便取一个时刻,让网络在这里预测的速度,去逼近这条直线的方向。"
|
|
541
474
|
}
|
|
542
475
|
]
|
|
543
476
|
},
|
|
@@ -583,7 +516,7 @@
|
|
|
583
516
|
"children": [
|
|
584
517
|
{
|
|
585
518
|
"type": "text",
|
|
586
|
-
"text": "
|
|
519
|
+
"text": "训练三万轮后,撒下两千个噪声粒子,沿学到的速度走,填满了整个标志。按位置给噪声染色,到了标志里颜色还连成一片,所以这个映射是连续的。"
|
|
587
520
|
}
|
|
588
521
|
]
|
|
589
522
|
},
|
|
@@ -629,7 +562,7 @@
|
|
|
629
562
|
"children": [
|
|
630
563
|
{
|
|
631
564
|
"type": "text",
|
|
632
|
-
"text": "
|
|
565
|
+
"text": "可网络的箭头始终对不齐那条线。随机配对的直线会交叉,在交叉处,网络只能学到各个方向的平均,所以真正走出来的路又弯了。"
|
|
633
566
|
}
|
|
634
567
|
]
|
|
635
568
|
},
|
|
@@ -721,7 +654,7 @@
|
|
|
721
654
|
"children": [
|
|
722
655
|
{
|
|
723
656
|
"type": "text",
|
|
724
|
-
"text": "
|
|
657
|
+
"text": "同样只走一步,第一个模型的粒子全挤在中间;拉直之后,一步就画出了标志。"
|
|
725
658
|
}
|
|
726
659
|
]
|
|
727
660
|
},
|
|
@@ -936,7 +869,7 @@
|
|
|
936
869
|
"content": [
|
|
937
870
|
{
|
|
938
871
|
"type": "text",
|
|
939
|
-
"text": "
|
|
872
|
+
"text": "生成一张图,就是把一团随机噪声推成图。每个噪声点沿着自己的路,走到图里属于它的位置。"
|
|
940
873
|
}
|
|
941
874
|
]
|
|
942
875
|
},
|
|
@@ -945,7 +878,7 @@
|
|
|
945
878
|
"content": [
|
|
946
879
|
{
|
|
947
880
|
"type": "text",
|
|
948
|
-
"text": "
|
|
881
|
+
"text": "网络不学整条路,只学每个位置、每个时刻该往哪走,也就是速度。速度就是路径的切线。"
|
|
949
882
|
}
|
|
950
883
|
]
|
|
951
884
|
},
|
|
@@ -954,7 +887,7 @@
|
|
|
954
887
|
"content": [
|
|
955
888
|
{
|
|
956
889
|
"type": "text",
|
|
957
|
-
"text": "
|
|
890
|
+
"text": "生成时,把时间切成几段,每段沿着速度走一步。可路是弯的,每一步都会偏出去一点。四步偏得明显,三十二步才贴住。"
|
|
958
891
|
}
|
|
959
892
|
]
|
|
960
893
|
},
|
|
@@ -972,7 +905,7 @@
|
|
|
972
905
|
"content": [
|
|
973
906
|
{
|
|
974
907
|
"type": "text",
|
|
975
|
-
"text": "Flow matching
|
|
908
|
+
"text": "Flow matching 的训练很简单。随便配一个噪声点和一个图上的点,连一条直线。在线上随便取一个时刻,让网络在这里预测的速度,去逼近这条直线的方向。"
|
|
976
909
|
}
|
|
977
910
|
]
|
|
978
911
|
},
|
|
@@ -981,7 +914,7 @@
|
|
|
981
914
|
"content": [
|
|
982
915
|
{
|
|
983
916
|
"type": "text",
|
|
984
|
-
"text": "
|
|
917
|
+
"text": "训练三万轮后,撒下两千个噪声粒子,沿学到的速度走,填满了整个标志。按位置给噪声染色,到了标志里颜色还连成一片,所以这个映射是连续的。"
|
|
985
918
|
}
|
|
986
919
|
]
|
|
987
920
|
},
|
|
@@ -990,7 +923,7 @@
|
|
|
990
923
|
"content": [
|
|
991
924
|
{
|
|
992
925
|
"type": "text",
|
|
993
|
-
"text": "
|
|
926
|
+
"text": "可网络的箭头始终对不齐那条线。随机配对的直线会交叉,在交叉处,网络只能学到各个方向的平均,所以真正走出来的路又弯了。"
|
|
994
927
|
}
|
|
995
928
|
]
|
|
996
929
|
},
|
|
@@ -1008,7 +941,7 @@
|
|
|
1008
941
|
"content": [
|
|
1009
942
|
{
|
|
1010
943
|
"type": "text",
|
|
1011
|
-
"text": "
|
|
944
|
+
"text": "同样只走一步,第一个模型的粒子全挤在中间;拉直之后,一步就画出了标志。"
|
|
1012
945
|
}
|
|
1013
946
|
]
|
|
1014
947
|
}
|