agentcut 0.2.15 → 0.2.16
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/dist/studio/chat.js +2 -1
- package/dist/studio/chat.js.map +1 -1
- package/dist/studio/host.d.ts +2 -0
- package/dist/studio/host.js +17 -5
- package/dist/studio/host.js.map +1 -1
- package/dist/studio/providers/http.d.ts +4 -0
- package/dist/studio/providers/http.js +23 -0
- package/dist/studio/providers/http.js.map +1 -0
- package/dist/studio/providers/settings.d.ts +20 -0
- package/dist/studio/providers/settings.js +84 -0
- package/dist/studio/providers/settings.js.map +1 -0
- package/dist/studio/schema.d.ts +1134 -36
- package/dist/studio/schema.js +10 -0
- package/dist/studio/schema.js.map +1 -1
- package/dist/studio/tools.js +11 -5
- package/dist/studio/tools.js.map +1 -1
- package/dist/studio/transcription/assemblyai.d.ts +11 -0
- package/dist/studio/transcription/assemblyai.js +123 -0
- package/dist/studio/transcription/assemblyai.js.map +1 -0
- package/dist/studio/transcription/local.d.ts +2 -0
- package/dist/studio/transcription/local.js +57 -0
- package/dist/studio/transcription/local.js.map +1 -0
- package/dist/studio/transcription/registry.d.ts +3 -0
- package/dist/studio/transcription/registry.js +5 -0
- package/dist/studio/transcription/registry.js.map +1 -0
- package/dist/studio/transcription/service.d.ts +130 -0
- package/dist/studio/transcription/service.js +235 -0
- package/dist/studio/transcription/service.js.map +1 -0
- package/dist/studio/transcription/types.d.ts +68 -0
- package/dist/studio/transcription/types.js +2 -0
- package/dist/studio/transcription/types.js.map +1 -0
- package/dist/studio/voice/audio.d.ts +1 -0
- package/dist/studio/voice/audio.js +18 -0
- package/dist/studio/voice/audio.js.map +1 -0
- package/dist/studio/voice/elevenlabs.d.ts +2 -0
- package/dist/studio/voice/elevenlabs.js +76 -0
- package/dist/studio/voice/elevenlabs.js.map +1 -0
- package/dist/studio/voice/gemini.d.ts +2 -0
- package/dist/studio/voice/gemini.js +95 -0
- package/dist/studio/voice/gemini.js.map +1 -0
- package/dist/studio/voice/openai.d.ts +2 -0
- package/dist/studio/voice/openai.js +75 -0
- package/dist/studio/voice/openai.js.map +1 -0
- package/dist/studio/voice/registry.d.ts +4 -0
- package/dist/studio/voice/registry.js +18 -0
- package/dist/studio/voice/registry.js.map +1 -0
- package/dist/studio/voice/types.d.ts +42 -0
- package/dist/studio/voice/types.js +2 -0
- package/dist/studio/voice/types.js.map +1 -0
- package/dist/studio/voice/xai.d.ts +2 -0
- package/dist/studio/voice/xai.js +80 -0
- package/dist/studio/voice/xai.js.map +1 -0
- package/dist/studio/voice.d.ts +32 -58
- package/dist/studio/voice.js +52 -308
- package/dist/studio/voice.js.map +1 -1
- package/package.json +1 -1
- package/template/components/CATALOG.md +2 -0
- package/template/components/STYLES.md +108 -7
- package/template/components/catalog.json +29 -0
- package/template/components/hook-card/HookCard.tsx +11 -5
- package/template/components/hook-card/meta.json +29 -0
- package/template/components/sticker-badge/StickerBadge.tsx +4 -2
- package/template/components/styles.json +354 -3
- package/template/runtime/composition.tsx +10 -2
- package/template/runtime/framing.ts +29 -0
- package/template/runtime/types.ts +4 -0
- package/template/skill/SKILL.md +13 -2
- package/template/skill/references/person-framing.md +13 -0
- package/template/skill/references/style-editing.md +20 -0
- package/template/skill/references/transcription.md +15 -0
|
@@ -521,15 +521,17 @@
|
|
|
521
521
|
"Captions carry the video: huge, uppercase, centered, one or two words at a time, the key word in yellow.",
|
|
522
522
|
"Explain every idea with a black full-frame Line Diagram (venn, one to many, growth, cycle, small vs big, stairs) while it is said.",
|
|
523
523
|
"Jump-cut punch-ins every ~2 s; no texture, no color besides the yellow.",
|
|
524
|
-
"Open with
|
|
524
|
+
"Open with a centered full-frame title in the line-art look. Use centered diagrams for the explanation, then return to the speaker. Do not use a split header for this opening."
|
|
525
525
|
],
|
|
526
526
|
"shots": [
|
|
527
527
|
{
|
|
528
528
|
"job": "Hook",
|
|
529
529
|
"component": "hook-card",
|
|
530
530
|
"props": {
|
|
531
|
-
"layout": "
|
|
532
|
-
"tagline": ""
|
|
531
|
+
"layout": "full",
|
|
532
|
+
"tagline": "",
|
|
533
|
+
"position": "center",
|
|
534
|
+
"align": "center"
|
|
533
535
|
}
|
|
534
536
|
},
|
|
535
537
|
{
|
|
@@ -758,5 +760,354 @@
|
|
|
758
760
|
}
|
|
759
761
|
}
|
|
760
762
|
]
|
|
763
|
+
},
|
|
764
|
+
{
|
|
765
|
+
"id": "signal",
|
|
766
|
+
"name": "Signal",
|
|
767
|
+
"description": "Direct business advice: charcoal, lime emphasis, bold captions and clear proof.",
|
|
768
|
+
"look": "clean",
|
|
769
|
+
"tokens": {
|
|
770
|
+
"look": "clean",
|
|
771
|
+
"font": "heavy,condensed",
|
|
772
|
+
"weight": "800",
|
|
773
|
+
"text": "#ffffff",
|
|
774
|
+
"accent": "#cbef59",
|
|
775
|
+
"onAccent": "#ffffff",
|
|
776
|
+
"panel": "#171b18",
|
|
777
|
+
"panelText": "#ffffff",
|
|
778
|
+
"line": "#455241",
|
|
779
|
+
"theme": "dark",
|
|
780
|
+
"backdrop": "#101511"
|
|
781
|
+
},
|
|
782
|
+
"camera": {
|
|
783
|
+
"mode": "still",
|
|
784
|
+
"zoom": 1,
|
|
785
|
+
"every": 150,
|
|
786
|
+
"shake": 0,
|
|
787
|
+
"grade": "contrast(1.02)"
|
|
788
|
+
},
|
|
789
|
+
"overlay": {
|
|
790
|
+
"kind": "none",
|
|
791
|
+
"amount": 0
|
|
792
|
+
},
|
|
793
|
+
"captions": {
|
|
794
|
+
"fontFamily": "Montserrat",
|
|
795
|
+
"fontWeight": 800,
|
|
796
|
+
"color": "#ffffff",
|
|
797
|
+
"highlight": "#cbef59",
|
|
798
|
+
"preset": "pop",
|
|
799
|
+
"position": "bottom",
|
|
800
|
+
"wordsPerLine": 3
|
|
801
|
+
},
|
|
802
|
+
"text": {
|
|
803
|
+
"fontFamily": "Montserrat",
|
|
804
|
+
"fontWeight": 800,
|
|
805
|
+
"color": "#ffffff"
|
|
806
|
+
},
|
|
807
|
+
"guide": [
|
|
808
|
+
"First transcribe the source with agentcut_transcribe. Review word timings and listen before removing dead air, repeated takes or filler words. Keep breaths, intentional pauses and the meaning of each sentence. Do not remove speech from transcript guesses alone.",
|
|
809
|
+
"Build the speech edit first with source trims and split clips. Close the resulting gaps and remap captions, B-roll and audio to the edited timeline. Keep the source files unchanged.",
|
|
810
|
+
"Review the picture about every 4–6 seconds for a useful change at a new idea, example or emphasis. Use a modest punch-in, wider reset or relevant B-roll only when it helps. Never insert cuts on a fixed five-second timer. The preset leaves the camera still so shot-level edits control the framing.",
|
|
811
|
+
"Use only supplied or approved B-roll and licensed audio. If footage is missing, use a useful text or diagram insert. Keep speech clear above music; measure the mix and check cuts for clicks. A color filter does not restore missing detail or upscale footage.",
|
|
812
|
+
"Lead with one useful claim, support it with an example, then give the next action. Keep the speaker prominent. Avoid decorative transitions and constant stickers.",
|
|
813
|
+
"Use short three-word caption groups, with lime on the key word. Use one stat counter for a real number and a compact list for actionable steps. Prefer straight cuts; return to the wider shot after an emphasis.",
|
|
814
|
+
"Keep skin tones neutral and the backdrop quiet. Use a low, steady music bed only if it helps; silence is valid for direct advice.",
|
|
815
|
+
"Open over full-frame footage with a compact left-aligned title card in the lower third (hook-card layout:card, position:lower, align:left). Keep it below the face and above the bottom safe zone; hide captions during this title."
|
|
816
|
+
],
|
|
817
|
+
"shots": [
|
|
818
|
+
{
|
|
819
|
+
"job": "Opening promise",
|
|
820
|
+
"component": "hook-card",
|
|
821
|
+
"props": {
|
|
822
|
+
"layout": "card",
|
|
823
|
+
"size": 15,
|
|
824
|
+
"position": "lower",
|
|
825
|
+
"align": "left"
|
|
826
|
+
}
|
|
827
|
+
},
|
|
828
|
+
{
|
|
829
|
+
"job": "Evidence",
|
|
830
|
+
"component": "stat-counter",
|
|
831
|
+
"props": {
|
|
832
|
+
"suffix": "%",
|
|
833
|
+
"label": "verified result"
|
|
834
|
+
}
|
|
835
|
+
},
|
|
836
|
+
{
|
|
837
|
+
"job": "Steps",
|
|
838
|
+
"component": "list-card",
|
|
839
|
+
"props": {
|
|
840
|
+
"layout": "card"
|
|
841
|
+
}
|
|
842
|
+
},
|
|
843
|
+
{
|
|
844
|
+
"job": "Next action",
|
|
845
|
+
"component": "word-pop",
|
|
846
|
+
"props": {
|
|
847
|
+
"position": "lower",
|
|
848
|
+
"tilt": false
|
|
849
|
+
}
|
|
850
|
+
}
|
|
851
|
+
]
|
|
852
|
+
},
|
|
853
|
+
{
|
|
854
|
+
"id": "studyroom",
|
|
855
|
+
"name": "Studyroom",
|
|
856
|
+
"description": "Warm lessons: cream pages, teal notes, calm type and clear chapter cards.",
|
|
857
|
+
"look": "clean",
|
|
858
|
+
"tokens": {
|
|
859
|
+
"look": "clean",
|
|
860
|
+
"font": "display,sans",
|
|
861
|
+
"weight": "600",
|
|
862
|
+
"text": "#ffffff",
|
|
863
|
+
"accent": "#1b766b",
|
|
864
|
+
"onAccent": "#202b28",
|
|
865
|
+
"panel": "#fff8e8",
|
|
866
|
+
"panelText": "#202b28",
|
|
867
|
+
"line": "#b7c9b6",
|
|
868
|
+
"theme": "light",
|
|
869
|
+
"backdrop": "#f2e8d4"
|
|
870
|
+
},
|
|
871
|
+
"camera": {
|
|
872
|
+
"mode": "still",
|
|
873
|
+
"zoom": 1,
|
|
874
|
+
"every": 150,
|
|
875
|
+
"shake": 0,
|
|
876
|
+
"grade": "saturate(0.95) sepia(0.04)"
|
|
877
|
+
},
|
|
878
|
+
"overlay": {
|
|
879
|
+
"kind": "none",
|
|
880
|
+
"amount": 0
|
|
881
|
+
},
|
|
882
|
+
"captions": {
|
|
883
|
+
"fontFamily": "Inter",
|
|
884
|
+
"fontWeight": 600,
|
|
885
|
+
"color": "#ffffff",
|
|
886
|
+
"highlight": "#1b766b",
|
|
887
|
+
"preset": "plain",
|
|
888
|
+
"position": "bottom",
|
|
889
|
+
"wordsPerLine": 6
|
|
890
|
+
},
|
|
891
|
+
"text": {
|
|
892
|
+
"fontFamily": "Inter",
|
|
893
|
+
"fontWeight": 600,
|
|
894
|
+
"color": "#ffffff"
|
|
895
|
+
},
|
|
896
|
+
"guide": [
|
|
897
|
+
"First transcribe the source with agentcut_transcribe. Review word timings and listen before removing dead air, repeated takes or filler words. Keep breaths, intentional pauses and the meaning of each sentence. Do not remove speech from transcript guesses alone.",
|
|
898
|
+
"Build the speech edit first with source trims and split clips. Close the resulting gaps and remap captions, B-roll and audio to the edited timeline. Keep the source files unchanged.",
|
|
899
|
+
"Review the picture about every 4–6 seconds for a useful change at a new idea, example or emphasis. Use a modest punch-in, wider reset or relevant B-roll only when it helps. Never insert cuts on a fixed five-second timer. The preset leaves the camera still so shot-level edits control the framing.",
|
|
900
|
+
"Use only supplied or approved B-roll and licensed audio. If footage is missing, use a useful text or diagram insert. Keep speech clear above music; measure the mix and check cuts for clicks. A color filter does not restore missing detail or upscale footage.",
|
|
901
|
+
"Organize the lesson into a question, a worked example and a short recap. Keep natural thinking pauses; remove only delays that do not help comprehension.",
|
|
902
|
+
"Use warm cream chapter cards with teal emphasis, a rounded speaker inset and six-word caption lines. Show screen recordings or diagrams during the explanation, then return to the face.",
|
|
903
|
+
"Apply only a slight warm grade after checking white balance. Use soft music and sparse, quiet reveal sounds. Let each example stay visible long enough to read."
|
|
904
|
+
],
|
|
905
|
+
"shots": [
|
|
906
|
+
{
|
|
907
|
+
"job": "Opening promise",
|
|
908
|
+
"component": "hook-card",
|
|
909
|
+
"props": {
|
|
910
|
+
"layout": "inset",
|
|
911
|
+
"size": 20
|
|
912
|
+
}
|
|
913
|
+
},
|
|
914
|
+
{
|
|
915
|
+
"job": "Chapter",
|
|
916
|
+
"component": "list-card",
|
|
917
|
+
"props": {
|
|
918
|
+
"layout": "card"
|
|
919
|
+
}
|
|
920
|
+
},
|
|
921
|
+
{
|
|
922
|
+
"job": "Worked example",
|
|
923
|
+
"component": "before-after",
|
|
924
|
+
"props": {
|
|
925
|
+
"layout": "full"
|
|
926
|
+
}
|
|
927
|
+
},
|
|
928
|
+
{
|
|
929
|
+
"job": "Takeaway",
|
|
930
|
+
"component": "quote-card",
|
|
931
|
+
"props": {
|
|
932
|
+
"layout": "card"
|
|
933
|
+
}
|
|
934
|
+
}
|
|
935
|
+
]
|
|
936
|
+
},
|
|
937
|
+
{
|
|
938
|
+
"id": "afterglow",
|
|
939
|
+
"name": "Afterglow",
|
|
940
|
+
"description": "Product stories: dark plum, copper titles, detail shots and restrained motion.",
|
|
941
|
+
"look": "clean",
|
|
942
|
+
"tokens": {
|
|
943
|
+
"look": "clean",
|
|
944
|
+
"font": "serif,display",
|
|
945
|
+
"weight": "400",
|
|
946
|
+
"text": "#fff5ec",
|
|
947
|
+
"accent": "#efa879",
|
|
948
|
+
"onAccent": "#ffffff",
|
|
949
|
+
"panel": "#211923",
|
|
950
|
+
"panelText": "#ffffff",
|
|
951
|
+
"line": "#71545d",
|
|
952
|
+
"theme": "dark",
|
|
953
|
+
"backdrop": "#130f16"
|
|
954
|
+
},
|
|
955
|
+
"camera": {
|
|
956
|
+
"mode": "still",
|
|
957
|
+
"zoom": 1,
|
|
958
|
+
"every": 150,
|
|
959
|
+
"shake": 0,
|
|
960
|
+
"grade": "contrast(1.08) saturate(0.9)"
|
|
961
|
+
},
|
|
962
|
+
"overlay": {
|
|
963
|
+
"kind": "vignette",
|
|
964
|
+
"amount": 0.18
|
|
965
|
+
},
|
|
966
|
+
"captions": {
|
|
967
|
+
"fontFamily": "Inter",
|
|
968
|
+
"fontWeight": 400,
|
|
969
|
+
"color": "#fff5ec",
|
|
970
|
+
"highlight": "#efa879",
|
|
971
|
+
"preset": "plain",
|
|
972
|
+
"position": "bottom",
|
|
973
|
+
"wordsPerLine": 6
|
|
974
|
+
},
|
|
975
|
+
"text": {
|
|
976
|
+
"fontFamily": "Inter",
|
|
977
|
+
"fontWeight": 400,
|
|
978
|
+
"color": "#fff5ec"
|
|
979
|
+
},
|
|
980
|
+
"guide": [
|
|
981
|
+
"First transcribe the source with agentcut_transcribe. Review word timings and listen before removing dead air, repeated takes or filler words. Keep breaths, intentional pauses and the meaning of each sentence. Do not remove speech from transcript guesses alone.",
|
|
982
|
+
"Build the speech edit first with source trims and split clips. Close the resulting gaps and remap captions, B-roll and audio to the edited timeline. Keep the source files unchanged.",
|
|
983
|
+
"Review the picture about every 4–6 seconds for a useful change at a new idea, example or emphasis. Use a modest punch-in, wider reset or relevant B-roll only when it helps. Never insert cuts on a fixed five-second timer. The preset leaves the camera still so shot-level edits control the framing.",
|
|
984
|
+
"Use only supplied or approved B-roll and licensed audio. If footage is missing, use a useful text or diagram insert. Keep speech clear above music; measure the mix and check cuts for clicks. A color filter does not restore missing detail or upscale footage.",
|
|
985
|
+
"Build a short product story: reveal, useful detail, demonstration, payoff. Keep the voice edit clean and use supplied close-ups to show the claim.",
|
|
986
|
+
"Alternate wide, detail and action shots at meaningful beats. Use match cuts where the source supports them. Use a brief light leak only at a scene change; avoid a transition on every cut.",
|
|
987
|
+
"Use large copper serif titles with generous empty space. Reserve captions for spoken explanation. Keep highlights below clipping and avoid heavy orange skin tones.",
|
|
988
|
+
"Place music hits and subtle source-backed Foley on actual actions. Do not add fake product sounds or claim speed ramps without source frames and an explicit timeline edit."
|
|
989
|
+
],
|
|
990
|
+
"shots": [
|
|
991
|
+
{
|
|
992
|
+
"job": "Opening promise",
|
|
993
|
+
"component": "hook-card",
|
|
994
|
+
"props": {
|
|
995
|
+
"layout": "frame",
|
|
996
|
+
"size": 20
|
|
997
|
+
}
|
|
998
|
+
},
|
|
999
|
+
{
|
|
1000
|
+
"job": "Reveal",
|
|
1001
|
+
"component": "light-leak",
|
|
1002
|
+
"props": {
|
|
1003
|
+
"strength": 0.35,
|
|
1004
|
+
"hueShift": 20
|
|
1005
|
+
}
|
|
1006
|
+
},
|
|
1007
|
+
{
|
|
1008
|
+
"job": "Product contrast",
|
|
1009
|
+
"component": "before-after",
|
|
1010
|
+
"props": {
|
|
1011
|
+
"layout": "full"
|
|
1012
|
+
}
|
|
1013
|
+
},
|
|
1014
|
+
{
|
|
1015
|
+
"job": "Payoff",
|
|
1016
|
+
"component": "word-pop",
|
|
1017
|
+
"props": {
|
|
1018
|
+
"position": "center",
|
|
1019
|
+
"tilt": false
|
|
1020
|
+
}
|
|
1021
|
+
}
|
|
1022
|
+
]
|
|
1023
|
+
},
|
|
1024
|
+
{
|
|
1025
|
+
"id": "fieldnote",
|
|
1026
|
+
"name": "Fieldnote",
|
|
1027
|
+
"description": "Documentary notes: ivory serif titles, rust markers, source cards and light grain.",
|
|
1028
|
+
"look": "clean",
|
|
1029
|
+
"tokens": {
|
|
1030
|
+
"look": "clean",
|
|
1031
|
+
"font": "serif,fraunces",
|
|
1032
|
+
"weight": "400",
|
|
1033
|
+
"text": "#ffffff",
|
|
1034
|
+
"accent": "#b94d32",
|
|
1035
|
+
"onAccent": "#24251f",
|
|
1036
|
+
"panel": "#f4f0e5",
|
|
1037
|
+
"panelText": "#24251f",
|
|
1038
|
+
"line": "#41463e",
|
|
1039
|
+
"theme": "light",
|
|
1040
|
+
"backdrop": "#e8e2d4"
|
|
1041
|
+
},
|
|
1042
|
+
"camera": {
|
|
1043
|
+
"mode": "still",
|
|
1044
|
+
"zoom": 1,
|
|
1045
|
+
"every": 150,
|
|
1046
|
+
"shake": 0,
|
|
1047
|
+
"grade": "contrast(1.02)"
|
|
1048
|
+
},
|
|
1049
|
+
"overlay": {
|
|
1050
|
+
"kind": "grain",
|
|
1051
|
+
"amount": 0.06
|
|
1052
|
+
},
|
|
1053
|
+
"captions": {
|
|
1054
|
+
"fontFamily": "Instrument Serif",
|
|
1055
|
+
"fontWeight": 400,
|
|
1056
|
+
"color": "#ffffff",
|
|
1057
|
+
"highlight": "#b94d32",
|
|
1058
|
+
"preset": "plain",
|
|
1059
|
+
"position": "bottom",
|
|
1060
|
+
"wordsPerLine": 6
|
|
1061
|
+
},
|
|
1062
|
+
"text": {
|
|
1063
|
+
"fontFamily": "Inter",
|
|
1064
|
+
"fontWeight": 400,
|
|
1065
|
+
"color": "#ffffff"
|
|
1066
|
+
},
|
|
1067
|
+
"guide": [
|
|
1068
|
+
"First transcribe the source with agentcut_transcribe. Review word timings and listen before removing dead air, repeated takes or filler words. Keep breaths, intentional pauses and the meaning of each sentence. Do not remove speech from transcript guesses alone.",
|
|
1069
|
+
"Build the speech edit first with source trims and split clips. Close the resulting gaps and remap captions, B-roll and audio to the edited timeline. Keep the source files unchanged.",
|
|
1070
|
+
"Review the picture about every 4–6 seconds for a useful change at a new idea, example or emphasis. Use a modest punch-in, wider reset or relevant B-roll only when it helps. Never insert cuts on a fixed five-second timer. The preset leaves the camera still so shot-level edits control the framing.",
|
|
1071
|
+
"Use only supplied or approved B-roll and licensed audio. If footage is missing, use a useful text or diagram insert. Keep speech clear above music; measure the mix and check cuts for clicks. A color filter does not restore missing detail or upscale footage.",
|
|
1072
|
+
"Build a setup, a question or conflict, and a supported resolution. Preserve pauses that carry emotion. Use captions as access to the speech, not as invented narration.",
|
|
1073
|
+
"Use ivory serif chapter titles, rust markers and source labels. Show evidence as supplied images or screen recordings, with a clear source caption. Never invent a quote, date, map or statistic.",
|
|
1074
|
+
"Use a wider speaker frame for context and a modest closer crop for a personal point. Hold on documents long enough to read. Use light grain and neutral color; preserve natural room tone.",
|
|
1075
|
+
"Use quote cards only for exact verified words. Keep music sparse under testimony, and finish with the answer to the opening question rather than an unrelated call to action.",
|
|
1076
|
+
"Open with a centered, full-frame ivory serif chapter card, not torn paper, tape or collage. Cut to the speaker once the question is clear. Keep rust markers small and evidence cards square and quiet."
|
|
1077
|
+
],
|
|
1078
|
+
"shots": [
|
|
1079
|
+
{
|
|
1080
|
+
"job": "Opening promise",
|
|
1081
|
+
"component": "hook-card",
|
|
1082
|
+
"props": {
|
|
1083
|
+
"layout": "full",
|
|
1084
|
+
"position": "center",
|
|
1085
|
+
"align": "center",
|
|
1086
|
+
"size": 18
|
|
1087
|
+
}
|
|
1088
|
+
},
|
|
1089
|
+
{
|
|
1090
|
+
"job": "Verified quote",
|
|
1091
|
+
"component": "quote-card",
|
|
1092
|
+
"props": {
|
|
1093
|
+
"layout": "full"
|
|
1094
|
+
}
|
|
1095
|
+
},
|
|
1096
|
+
{
|
|
1097
|
+
"job": "Evidence summary",
|
|
1098
|
+
"component": "list-card",
|
|
1099
|
+
"props": {
|
|
1100
|
+
"layout": "full"
|
|
1101
|
+
}
|
|
1102
|
+
},
|
|
1103
|
+
{
|
|
1104
|
+
"job": "Chapter marker",
|
|
1105
|
+
"component": "word-pop",
|
|
1106
|
+
"props": {
|
|
1107
|
+
"position": "lower",
|
|
1108
|
+
"tilt": false
|
|
1109
|
+
}
|
|
1110
|
+
}
|
|
1111
|
+
]
|
|
761
1112
|
}
|
|
762
1113
|
]
|
|
@@ -14,6 +14,7 @@ import {
|
|
|
14
14
|
useVideoConfig,
|
|
15
15
|
} from "remotion";
|
|
16
16
|
import { Audio, Video } from "@remotion/media";
|
|
17
|
+
import { framingCropAt } from "./framing";
|
|
17
18
|
// The Agentcut component library, provided by the studio page (and by a copy in
|
|
18
19
|
// .agentcut/components for local renders).
|
|
19
20
|
import * as ComponentLibrary from "agentcut/components";
|
|
@@ -362,13 +363,19 @@ function VideoClipView({
|
|
|
362
363
|
const style = useClipStyle(clip);
|
|
363
364
|
const frame = useCurrentFrame();
|
|
364
365
|
// The project style's camera (punch-ins, drift, …) moves the footage inside its box.
|
|
366
|
+
const crop = clip.framing ? framingCropAt(clip.framing, (clip.trimStart ?? 0) / ctx.project.fps + frame / ctx.project.fps * (clip.playbackRate ?? 1)) : null;
|
|
365
367
|
const camera =
|
|
366
368
|
ctx.project.style && clip.camera !== "none" && styleLayer.styleCamera
|
|
367
369
|
? styleLayer.styleCamera(ctx.project.style, { frame, durationInFrames: clip.duration, width: ctx.width, height: ctx.height })
|
|
368
370
|
: null;
|
|
369
371
|
return (
|
|
370
372
|
<div style={{ ...boxStyle(clip.box), ...style, overflow: "hidden", borderRadius: clip.radius }}>
|
|
371
|
-
<div style={{ position: "absolute", inset: 0, ...camera }}>
|
|
373
|
+
<div style={{ position: "absolute", inset: 0, ...(clip.framing ? { filter: camera?.filter } : camera) }}>
|
|
374
|
+
<div style={crop ? {
|
|
375
|
+
position: "absolute", left: `${-crop.x / crop.width * 100}%`,
|
|
376
|
+
top: `${-crop.y / crop.height * 100}%`,
|
|
377
|
+
width: `${100 / crop.width}%`, height: `${100 / crop.height}%`,
|
|
378
|
+
} : { position: "absolute", inset: 0 }}>
|
|
372
379
|
<Video
|
|
373
380
|
src={asset(clip.src)}
|
|
374
381
|
trimBefore={clip.trimStart && clip.trimStart > 0 ? clip.trimStart : undefined}
|
|
@@ -376,11 +383,12 @@ function VideoClipView({
|
|
|
376
383
|
muted={trackMuted || clip.muted}
|
|
377
384
|
playbackRate={clip.playbackRate}
|
|
378
385
|
// @remotion/media draws the frame itself: fit has to be its prop (CSS object-fit is ignored).
|
|
379
|
-
objectFit={clip.fit ?? "cover"}
|
|
386
|
+
objectFit={clip.framing ? "fill" : clip.fit ?? "cover"}
|
|
380
387
|
style={{ width: "100%", height: "100%", filter: gradeFilter(clip.grade) }}
|
|
381
388
|
/>
|
|
382
389
|
{clip.cursor ? <CursorOverlay cursor={clip.cursor} clip={clip} /> : null}
|
|
383
390
|
</div>
|
|
391
|
+
</div>
|
|
384
392
|
</div>
|
|
385
393
|
);
|
|
386
394
|
}
|
|
@@ -0,0 +1,29 @@
|
|
|
1
|
+
import type { Box, FramingKeyframe } from "./types";
|
|
2
|
+
|
|
3
|
+
/** Stateless source-time interpolation: identical for seeking, preview and every export frame. */
|
|
4
|
+
export function framingCropAt(
|
|
5
|
+
framing: { crop: Box; path?: FramingKeyframe[] },
|
|
6
|
+
seconds: number,
|
|
7
|
+
): Box {
|
|
8
|
+
const path = framing.path;
|
|
9
|
+
if (!path?.length || !Number.isFinite(seconds)) return framing.crop;
|
|
10
|
+
if (seconds <= path[0].seconds) return path[0].crop;
|
|
11
|
+
if (seconds >= path[path.length - 1].seconds)
|
|
12
|
+
return path[path.length - 1].crop;
|
|
13
|
+
let low = 0,
|
|
14
|
+
high = path.length - 1;
|
|
15
|
+
while (high - low > 1) {
|
|
16
|
+
const mid = (low + high) >> 1;
|
|
17
|
+
if (path[mid].seconds <= seconds) low = mid;
|
|
18
|
+
else high = mid;
|
|
19
|
+
}
|
|
20
|
+
const a = path[low],
|
|
21
|
+
b = path[high];
|
|
22
|
+
const progress = (seconds - a.seconds) / (b.seconds - a.seconds);
|
|
23
|
+
return {
|
|
24
|
+
x: a.crop.x + (b.crop.x - a.crop.x) * progress,
|
|
25
|
+
y: a.crop.y + (b.crop.y - a.crop.y) * progress,
|
|
26
|
+
width: a.crop.width + (b.crop.width - a.crop.width) * progress,
|
|
27
|
+
height: a.crop.height + (b.crop.height - a.crop.height) * progress,
|
|
28
|
+
};
|
|
29
|
+
}
|
|
@@ -93,6 +93,8 @@ export interface ColorGrade {
|
|
|
93
93
|
hue?: number;
|
|
94
94
|
}
|
|
95
95
|
|
|
96
|
+
export interface FramingKeyframe { seconds: number; crop: Box }
|
|
97
|
+
|
|
96
98
|
export interface VideoClipFields {
|
|
97
99
|
src: string;
|
|
98
100
|
trimStart?: number;
|
|
@@ -108,6 +110,8 @@ export interface VideoClipFields {
|
|
|
108
110
|
cursor?: VideoCursor;
|
|
109
111
|
/** "none" keeps the project style's camera moves off this clip. */
|
|
110
112
|
camera?: "style" | "none";
|
|
113
|
+
/** Stable crop in upright source fractions, shared by preview and export. */
|
|
114
|
+
framing?: { crop: Box; path?: FramingKeyframe[]; before?: Box | null; beforePath?: FramingKeyframe[] };
|
|
111
115
|
}
|
|
112
116
|
|
|
113
117
|
export interface VideoCursor {
|
package/template/skill/SKILL.md
CHANGED
|
@@ -41,7 +41,8 @@ projects/<slug>/
|
|
|
41
41
|
look, fonts, palette, text animation, camera moves, captions and texture. Read
|
|
42
42
|
`.agentcut/components/STYLES.md` when the project has a `"style"`, when the user names a style
|
|
43
43
|
("use the Snipwork style", "make it attention grabbing"), or when a message asks you to restyle.
|
|
44
|
-
Applying one =
|
|
44
|
+
Applying one = read `references/style-editing.md`, then follow "Applying a style" there.
|
|
45
|
+
A style sets visual defaults. Speech cleanup, B-roll and shot-level framing require timeline edits. With a style
|
|
45
46
|
set, leave style-driven props out of clips, and make your own scene files match its tokens.
|
|
46
47
|
- **Video style presets** are recipes for whole videos that aren't talking heads (music and lyric videos, …), kept in
|
|
47
48
|
`library/presets/<id>/` with a PROMPT.md. When a message names one, or asks to save the open video's style as a
|
|
@@ -49,6 +50,14 @@ projects/<slug>/
|
|
|
49
50
|
|
|
50
51
|
## Loop
|
|
51
52
|
|
|
53
|
+
For an explicit request such as "center me in the frame", use `agentcut_frame_person`.
|
|
54
|
+
Read `references/person-framing.md` first. Do not install a detector or write a face-detection script.
|
|
55
|
+
Do not center footage just because the output format changes.
|
|
56
|
+
|
|
57
|
+
For speech recognition use `agentcut_transcribe`, with provider discovery from
|
|
58
|
+
`agentcut_transcription_options`. Read `references/transcription.md` before choosing a provider.
|
|
59
|
+
Use the selected preference unless the user requests an override. Keys stay in settings.
|
|
60
|
+
|
|
52
61
|
1. `agentcut_status` — confirms the page is connected, which project is open, and any
|
|
53
62
|
build or validation errors. Use `agentcut_create_project` for a new video (it opens it).
|
|
54
63
|
2. **Plan as a cue sheet** in `notes.md`: beats with frame ranges. When there's a voiceover, the voice
|
|
@@ -162,7 +171,9 @@ scripts from the project folder (e.g. composing music or fetching CC0 sounds, se
|
|
|
162
171
|
| `agentcut_capture_frames` | look at specific frames (same renderer as export); `region` + `scale: 2` for a sharp crop |
|
|
163
172
|
| `agentcut_contact_sheet` | stills sheet of the whole video before export |
|
|
164
173
|
| `agentcut_media_info` | exact durations (frames) for clips |
|
|
165
|
-
| `agentcut_transcribe` | word timings
|
|
174
|
+
| `agentcut_transcribe` | provider-based transcript and word timings (see `references/transcription.md`) |
|
|
175
|
+
| `agentcut_transcription_options` / `_cancel_transcription` | configured capabilities, progress and cancellation |
|
|
176
|
+
| `agentcut_frame_person` | explicit local face centering or crop reset (see `references/person-framing.md`) |
|
|
166
177
|
| `agentcut_preview_check` | does the live preview keep up? dropped frames per clip (see `references/performance.md`) |
|
|
167
178
|
| `agentcut_import_reference` | Style Spy: a reference video plus sampled frames (see `references/style-spy.md`) |
|
|
168
179
|
| `agentcut_voice_options` / `_generate_voice` | voice-overs with the user's own provider (see `references/sound.md`) |
|
|
@@ -0,0 +1,13 @@
|
|
|
1
|
+
# Person centering and following
|
|
2
|
+
|
|
3
|
+
Use agentcut_frame_person for an explicit request to center or keep a visible speaker centered. Do not install detection packages, upload frames, use image APIs or guess a crop from a screenshot. An aspect-ratio change alone does not authorize framing.
|
|
4
|
+
|
|
5
|
+
1. Call agentcut_status. Pass the target variant and video clip IDs. The tool defaults to selected clips, or the sole visible, unlocked video clip. With several candidates, select the intended clip or ask the user. Maximum eight clips; all must succeed before any path is saved.
|
|
6
|
+
2. For ordinary keep-centered requests, use action:center and mode:follow (the default). The full displayed source range is sampled at no more than 0.5-second source intervals, including the first and last source frame. Trim, playback speed, source display dimensions, rotation and pixel aspect ratio are respected. Up to 10 minutes of source time per clip and 1,800 samples per call are supported. Split longer shots or process fewer clips; the tool does not silently inspect only part of a clip. mode:static explicitly requests one fixed crop from nine samples.
|
|
7
|
+
3. Follow mode retains one spatially continuous face, checks position/size changes, applies a dead zone and low-pass smoothing, and bounds the crop inside the source. It holds the last crop during detection loss of up to 1.5 seconds. It rejects multiple faces, long loss, weak coverage, large picture changes, abrupt subject changes and crop jumps. These return structured status and next steps, without applying a partial path. Ask for a clearer or shorter shot, or a wider format. Do not switch people or guess after failure.
|
|
8
|
+
4. Results report coverage, held samples, keyframe count and boundsLimited. Narrow output or source edges can limit ideal centering and headroom; no detail is restored. Source frames remain local in a CPU worker; the conversational agent still uses tokens, but detection makes no LLM/image API calls. Model/runtime files come from Agentcut.
|
|
9
|
+
5. Preview and export evaluate the same persisted crop path at source time: trimStart/fps + localFrame/fps * playbackRate. Seeking is deterministic. Timing, clip box, grade and other variants remain unchanged. Style camera motion is suspended while framing is set. The crop/path replaces cover/contain fitting inside the existing box. Original project or library videos are supported; scene-embedded video and cursor overlays are not.
|
|
10
|
+
6. Review captures at the start, moving sections and end in the target variant. Detection is sampled, not frame-by-frame, and does not recognize identity. It can miss a short cut, face or fast movement between samples. Large picture/face jumps are treated as possible cuts, not guaranteed scene detection. The tool asks for shot splits rather than selecting an identity or crossing an uncertain cut.
|
|
11
|
+
7. Reset with action:reset and the same clips/variant. It restores prior crop and path overrides and preserves later unrelated edits. Normal studio Undo restores the previous timeline. After an unconfirmed save or conflict, inspect the timeline before retrying. Update and restart the helper/runtime before using this new field; do not restart during the user's live work.
|
|
12
|
+
|
|
13
|
+
Persisted field: variants.<variant>.framing contains crop (upright source fractions), optional path entries {seconds,crop} at absolute source times, and before/beforePath reset metadata. Prefer the tool over hand edits. Do not copy a path to another source/aspect ratio without analysis. Later trim changes outside the analyzed source range hold the end crop; rerun the tool to analyze any newly exposed source.
|
|
@@ -0,0 +1,20 @@
|
|
|
1
|
+
# Style editing workflow
|
|
2
|
+
|
|
3
|
+
A selected style supplies visual defaults and an editing guide. Selection alone does not clean speech, find footage, improve source resolution or mix audio. Carry out these steps when the user sends an editing request.
|
|
4
|
+
|
|
5
|
+
1. Inspect the source and the current timeline. Read the selected entry in .agentcut/components/STYLES.md. Preserve user edits. Record a short cue sheet in notes.md.
|
|
6
|
+
2. Transcribe with agentcut_transcribe. Review timed words and listen to candidate cuts. Remove verified repeated takes, unnecessary silence and filler words only where the sentence still sounds natural. Keep meaningful pauses and breaths. If timing or listening is unavailable, do not guess the cuts; report the limit.
|
|
7
|
+
3. Assemble speech first. Split and trim source clips, close gaps, and keep source files unchanged. Rebuild caption timings against the edited sequence. Move related graphics, B-roll and audio with their speech. Check every join for lost syllables and audio clicks.
|
|
8
|
+
4. Review framing about every 4–6 seconds. Choose a new crop, wider reset or relevant B-roll at a change of idea, example or emphasis. This is a review interval, not a fixed cut interval. Preserve face and eye lines in each variant. Use supported clip boxes or an explicit scene transform; do not invent a zoom property. Keep camera:none on clips if a style camera would add an unwanted repeating move. The four new styles use a still camera by default.
|
|
9
|
+
5. Add supplied or approved B-roll to explain the words. Use source labels on evidence. If needed footage is missing, use a useful card or diagram and report the missing asset. Do not invent quotes, statistics or product details.
|
|
10
|
+
6. Apply the palette, type and restrained grade. Check skin tones, highlight clipping and text contrast. A filter cannot restore focus or upscale source detail. Keep music below speech, use sparse sounds on visible cues, and follow references/sound.md for measurement. Apply EQ or compression only through supported tools or a documented audio-processing step; a preset name does not apply them.
|
|
11
|
+
7. Capture entrances, holds and exits in each size. Use a contact sheet for pacing. Listen to the final mix. Check caption sync after all cuts. Report what changed and any source or tool limits.
|
|
12
|
+
|
|
13
|
+
## Research behind the four directions
|
|
14
|
+
|
|
15
|
+
These are original Agentcut styles inspired by editing principles. They are not creator replicas, endorsements or a ranking of current trends. Palettes, layouts and preview choreography are our design choices.
|
|
16
|
+
|
|
17
|
+
- Signal: direct business advice, evidence and a next action. Alex Hormozi's June 2024 note favors useful business content and less distracting production: https://www.acquisition.com/mozi-money-minute/i-made-a-youtube-video . His official workbook describes hook, retain and reward: https://shop.acquisition.com/products/workbook-3-pack-digital . This supports a clear message, not constant animated captions.
|
|
18
|
+
- Studyroom: lesson assembly, relevant supporting visuals, then sound and color polish. Ali Abdaal explains this three-stage workflow: https://aliabdaal.com/youtube/the-ultimate-guide-to-youtube/ . Preserve useful pauses rather than applying blanket silence removal.
|
|
19
|
+
- Afterglow: a product story with detail footage, a measured grade and music-led scene changes. Daniel Schiffer's own cooking B-roll editing breakdown is a visual reference: https://www.youtube.com/watch?v=qSMi0ygL--E . Actual footage and timeline edits are required; a look preset alone does not create that sequence.
|
|
20
|
+
- Fieldnote: documentary setup, question and resolution with traceable evidence. Colin and Samir discuss audience, story planning and a three-act structure in their first-person interview: https://www.musicbed.com/articles/filmmaking/youtube/the-biggest-mistake-filmmakers-make-with-colin-and-samir/ . Serif type and rust source markers are Agentcut design choices.
|
|
@@ -0,0 +1,15 @@
|
|
|
1
|
+
# Transcription providers
|
|
2
|
+
|
|
3
|
+
Use the built-in operation. Do not install a provider SDK, ask for keys in chat, or write upload scripts.
|
|
4
|
+
|
|
5
|
+
- `agentcut_transcription_options` lists preferred provider, configured status, models, capabilities and jobs, never credentials. Local Whisper is the initial default. Adding a cloud key does not select it. The user selects a preference under Transcription in either editor view.
|
|
6
|
+
- `agentcut_transcribe` accepts `path`, optional `project`, `provider`, `model`, `language`, `speakerLabels` and a unique optional `job` ID. Use the preference unless the user explicitly asks for an override, such as `provider: "local"`. Never silently switch after an error. For unconfigured providers ask the user to open settings; never ask them to paste a key in chat.
|
|
7
|
+
- Local Whisper retains tiny/base/small models and word timings. Language defaults to en; it needs an explicit language code and does not supply speaker labels. Filler recognition is not guaranteed. Its files stay local, though model downloads may be needed.
|
|
8
|
+
- AssemblyAI uses the user's API key in the helper. Audio is extracted locally, then uploaded to AssemblyAI. Provider charges and provider data handling apply. It uses Universal-2 and always requests `disfluencies: true` for editing. That model's filler support is English only; other languages and automatic language detection do not guarantee filler recognition. Speaker labels are optional. A key does not imply permission to run unsolicited paid test jobs.
|
|
9
|
+
- Results use seconds (`start`, `end`) and project frames (`fromFrame`, `toFrame`) from the original source, not edited timeline positions. Preserve trims and map words through cuts. Results include provider/model/options/source provenance, a cache key, segments and speaker labels when available. Cached results are separated by source content, provider, options and fps. Raw recognized filler words and their timings are retained. A transcript that omits fillers is not evidence that the audio was edited.
|
|
10
|
+
- Pass a unique `job` ID if you need progress or cancellation. Discover stages with `agentcut_transcription_options`, cancel with `agentcut_cancel_transcription`. Transcription settings also shows active jobs and Cancel. Cancellation stops local processing or cloud polling; remote work and charges may already have begun. No other provider is used as fallback.
|
|
11
|
+
- Listen before making silence/filler cuts. ASR timings and words can be wrong, especially with music, overlap, accents or poor audio. Do not equate a cleaner text result with edited speech.
|
|
12
|
+
|
|
13
|
+
Implementation extension points are separate: `src/studio/transcription/registry.ts` for recognition and `src/studio/voice/registry.ts` for synthesis. An adapter owns metadata, capabilities and provider calls. Both use helper-only provider settings/secret storage. Adding a synthesis provider does not make it a transcription provider.
|
|
14
|
+
|
|
15
|
+
Primary references: https://www.assemblyai.com/docs/pre-recorded-audio/getting-started/transcribe-an-audio-file , https://www.assemblyai.com/docs/pre-recorded-audio/include-filler-words , https://www.assemblyai.com/docs/pre-recorded-audio/select-the-speech-model .
|