@aicopycoders/exodus 2026.11.700-beta.1 → 2026.11.1000-beta.2
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 +357 -0
- package/assets/skills/exodus-video/SKILL.md +309 -1
- package/assets/skills/exodus-video/scripts/first-cut-text.d.mts +4 -1
- package/assets/skills/exodus-video/scripts/first-cut-text.mjs +20 -2
- package/assets/skills/exodus-video/scripts/first-cut.d.mts +80 -4
- package/assets/skills/exodus-video/scripts/first-cut.mjs +1248 -132
- package/assets/skills/swipe/SKILL.md +11 -2
- package/dist/commands/video-customer.d.ts +1 -1
- package/dist/commands/video-customer.js +7 -0
- package/dist/commands/video.d.ts +160 -2
- package/dist/commands/video.js +1633 -43
- package/dist/commands/workflow.d.ts +3 -3
- package/dist/commands/workflow.js +20 -2
- package/dist/lib/templateBrief.d.ts +7 -0
- package/dist/lib/templateBrief.js +9 -0
- package/package.json +1 -1
package/README.md
CHANGED
|
@@ -40,12 +40,369 @@ npx -y @aicopycoders/exodus@beta doctor
|
|
|
40
40
|
`doctor` diagnoses local and dashboard-side issues in one shot — keys, workspace
|
|
41
41
|
health, CLI currency — and tells you exactly what to fix.
|
|
42
42
|
|
|
43
|
+
### Template video inputs
|
|
44
|
+
|
|
45
|
+
Native-audio template final cuts use the measured, post-trim clip length when an
|
|
46
|
+
approved picture reservation is longer than the available footage. Every spoken
|
|
47
|
+
word and required post-word action tail must still fit; unavailable speech/action
|
|
48
|
+
is refused rather than freeze-filled. Voice-first and ordinary script-door cuts
|
|
49
|
+
keep their existing timing. After the worker update, **Cut again** can reuse the
|
|
50
|
+
existing clips for this reservation-only failure; it does not require generating
|
|
51
|
+
new clips.
|
|
52
|
+
|
|
53
|
+
Storyboard caption and overlay boxes use normalized `[x, y, width, height]`.
|
|
54
|
+
The first-cut text adapter converts them to the renderer's
|
|
55
|
+
`[left, top, right, bottom]`; source-ad reference boxes already use that renderer
|
|
56
|
+
contract and are preserved. Invalid geometry still holds the finish without
|
|
57
|
+
clamping. After the worker update, **Cut again** can retry a caption-geometry
|
|
58
|
+
adapter failure using the existing clips; no clip regeneration is needed.
|
|
59
|
+
|
|
60
|
+
#### V2 input assistance and script review
|
|
61
|
+
|
|
62
|
+
Template input assistance is **not execution automation**:
|
|
63
|
+
|
|
64
|
+
- `--input-mode automatic` asks the server to resolve supported, unfilled choices.
|
|
65
|
+
- `--input-mode manual` exposes supported creative controls through the template's
|
|
66
|
+
published capabilities and `--template-brief`.
|
|
67
|
+
- `--input-mode hybrid` uses the same controls as Manual; blank supported fields
|
|
68
|
+
are automatic. There is no per-field checkbox or separate override map.
|
|
69
|
+
|
|
70
|
+
All modes retain explicit creative pins, model, voice route, account voices, saved
|
|
71
|
+
references and product selection. Server-owned locks still define the format:
|
|
72
|
+
Skeleton remains Skeleton, and a song-led format remains song-led. Unsupported or
|
|
73
|
+
incompatible values are refused, not silently cleared. No mode grants identity,
|
|
74
|
+
product facts, consent, soundtrack rights or permission to spend. V2 input controls
|
|
75
|
+
do not accept `--one-shot`; storyboard review remains required.
|
|
76
|
+
|
|
77
|
+
`--structure <key>` selects one of that template's published story structures.
|
|
78
|
+
`--reference <key>` selects reference-ad inspiration only, never production
|
|
79
|
+
conditioning or voice rights. Use authenticated discovery instead of guessing
|
|
80
|
+
keys or borrowing a different template's structures.
|
|
81
|
+
Reference examples can include advertiser references and historical generated
|
|
82
|
+
demos. Discovery prints each supplied source label and URL; the collection itself
|
|
83
|
+
does not certify that every example is an original ad or was made by us.
|
|
84
|
+
|
|
85
|
+
`--subject product|service|non-product` records an explicit subject. Absence remains
|
|
86
|
+
unknown, not an inferred product/service choice. Service and non-product briefs
|
|
87
|
+
must not select a saved product. Supply factual evidence in the source text;
|
|
88
|
+
subject selection does not grant new facts or claims.
|
|
89
|
+
|
|
90
|
+
Choose one explicit script entry path:
|
|
91
|
+
|
|
92
|
+
- **Write from a brief:** `--describe --script brief.txt`.
|
|
93
|
+
- **Adapt pasted copy:** `--adapt "<what may change>" --script original.txt`.
|
|
94
|
+
The original file and adaptation intent are retained byte for byte. An adaptation
|
|
95
|
+
request cannot supply new product claims or inferred identities/consent.
|
|
96
|
+
- **Intentionally keep original copy:** add `--keep-original` to the `--adapt`
|
|
97
|
+
command. This requests guarded review of the exact original instead of generating
|
|
98
|
+
changes. Any genuinely billable approval still needs a phase quote. The output
|
|
99
|
+
explicitly says no adaptation was applied when the text is unchanged.
|
|
100
|
+
|
|
101
|
+
Mode/structure/manual-control flags alone never request a new script. A new V2
|
|
102
|
+
draft needs an explicit entry path above; no adaptation intent is inferred.
|
|
103
|
+
|
|
104
|
+
A new V2 write always returns a script-review stop before storyboard or video
|
|
105
|
+
generation. This is **not a promise of free preparation**: writer and claim-checking
|
|
106
|
+
work can be paid, and input assistance itself never approves that spending.
|
|
107
|
+
The human stop output displays the original pasted text, explicit intent and full
|
|
108
|
+
resulting script for comparison. `--json` retains the exact strings and the
|
|
109
|
+
server-issued approval. Keep the saved JSON unchanged and edit a separate script
|
|
110
|
+
file.
|
|
111
|
+
Extract with `jq -j '.script'`, not `jq -r`, when preparing an exact-text resume:
|
|
112
|
+
`-r` appends a newline that was not necessarily present in the reviewed result.
|
|
113
|
+
|
|
114
|
+
`--template-brief '<json>'` or `--template-brief @brief.json` supplies the existing
|
|
115
|
+
`{ "cards": { "<layer>": "<card>" }, "settings": { "<name>": <primitive> } }`
|
|
116
|
+
creative pins. The CLI checks only this transport shape; the server resolves
|
|
117
|
+
actual capabilities, locks and dependent choices. Dedicated `--requested-seconds`
|
|
118
|
+
and `--pictures-per-scene` flags override matching settings. Required `--style`
|
|
119
|
+
selects the launch look; matching the authenticated template default does not
|
|
120
|
+
automatically add a `cards.style` override (locked looks refuse even same-value
|
|
121
|
+
card pins). An explicit supplied or saved brief-style pin remains explicit for
|
|
122
|
+
server validation. A resumed brief merges supplied fields with saved pins, so
|
|
123
|
+
changing input mode never discards omitted choices.
|
|
124
|
+
|
|
125
|
+
Supported `--song-controls <json|@file>` and `--mechanism-plan <json|@file>` are
|
|
126
|
+
existing server-validated JSON objects, not arbitrary extra facts or engine
|
|
127
|
+
parameters. Explicit false music, frame/pacing shots and these sound/story controls
|
|
128
|
+
are retained on V2 stop/resume along with direction. A blank string in a creative
|
|
129
|
+
card/setting means unfilled: it removes that pin before server resolution. Nonblank
|
|
130
|
+
explicit values survive mode changes unchanged.
|
|
131
|
+
Song controls are offered only when the selected server template advertises
|
|
132
|
+
`templateCapabilities.songControls: true`. The existing fields are `genre`,
|
|
133
|
+
`mood`, `bpm`, `vocal`, `instruments`, and `durationSec`; for example,
|
|
134
|
+
`--song-controls '{"genre":"acoustic pop","mood":"hopeful"}'`. Absent capability
|
|
135
|
+
means unsupported, not a guessed format-ID exception. Nonempty incompatible pins
|
|
136
|
+
are held, not discarded; `--song-controls '{}'` (or all-blank text fields) explicitly
|
|
137
|
+
clears them for automatic resolution. Musical instructions do not grant soundtrack
|
|
138
|
+
or voice rights.
|
|
139
|
+
|
|
140
|
+
Discovery returns the authenticated server's `templateExperience` and
|
|
141
|
+
`templateCapabilities`; there is no bundled fallback catalog. Before quoting, the
|
|
142
|
+
CLI reads selected-draft input options using the exact intent/brief/route. When
|
|
143
|
+
published, safe `intentResolution` is retained separately in quote/review JSON and
|
|
144
|
+
shows `effectiveBrief`, explained holds, and `template-default` / `resolved` /
|
|
145
|
+
`pinned` provenance. Effective defaults never become new raw pins. Held or stale
|
|
146
|
+
resolution cannot proceed to a quote. If the server omits this optional view, the
|
|
147
|
+
CLI says provenance was not published instead of pretending defaults were
|
|
148
|
+
intelligently chosen; authoritative phase admission still applies.
|
|
149
|
+
|
|
150
|
+
Resuming a V2 stop retains its original `templateIntent`, subject, source, brief,
|
|
151
|
+
product, model, voice route, saved elements, voice map and current creative context,
|
|
152
|
+
including declared source-record and semantic adaptation-plan JSON when supplied.
|
|
153
|
+
A changed script, mode, structure/reference, creative pin, model, sound or voice
|
|
154
|
+
route needs another explicit `--stop-at-script --json` review before starting;
|
|
155
|
+
use the newly returned response and separately authorize its phase quote.
|
|
156
|
+
Changing product requires a fresh write using the original source, without
|
|
157
|
+
`--resume-script`. An older response without V2 original-input context cannot be
|
|
158
|
+
upgraded by attaching V2 flags to its receipt. Server verification remains
|
|
159
|
+
authoritative for workspace/template/product/context binding; editing saved JSON
|
|
160
|
+
does not confer approval.
|
|
161
|
+
|
|
162
|
+
#### V2 phase quote → exact script review → storyboard preparation
|
|
163
|
+
|
|
164
|
+
`--quote-preparation --json` requests a free, server-issued quote for the exact
|
|
165
|
+
current phase and inputs. It performs no writer, audio, image or video preparation.
|
|
166
|
+
The saved CLI envelope contains `preparationQuote`, stable `requestId`, and the
|
|
167
|
+
server-returned canonical action `input`. Review `estimatedUsd` (an approximate
|
|
168
|
+
point estimate, or `null` when actual cost is unknown), `maximumUsd` (the enforceable
|
|
169
|
+
allowance for this phase only), `scope`, and `expiresAt`. An unknown estimate is
|
|
170
|
+
neither zero nor the maximum; none of these values is a whole-video price.
|
|
171
|
+
The HTTP quote's `canonicalInput` uses the same existing server parsers as paid
|
|
172
|
+
preparation. The CLI stores that representation rather than independently trimming
|
|
173
|
+
Song or mechanism fields. Equivalent padded controls can continue against the same
|
|
174
|
+
canonical context; genuinely changed controls require fresh script review. A paid
|
|
175
|
+
review response must match the quoted canonical controls before it becomes a
|
|
176
|
+
resumable response. Original source bytes and explicit input intent are not
|
|
177
|
+
normalized or replaced by this process.
|
|
178
|
+
|
|
179
|
+
Only **`--authorize-preparation <saved-quote.json>`** accepts that quote's maximum
|
|
180
|
+
for its exact current phase/input. Before submitting paid work the CLI re-reads
|
|
181
|
+
the **same** quote operation with the same request ID and requires unchanged
|
|
182
|
+
authority. Missing/held/expired/replaced quotes and changed draft inputs cannot
|
|
183
|
+
fall through to the legacy paid route. Do not edit the saved file. For an ambiguous
|
|
184
|
+
failure, keep that same file and recover the same operation; the CLI never silently
|
|
185
|
+
creates new authority or retries paid preparation. A server cost/provider/worker
|
|
186
|
+
hold is a dependency to resolve, not permission to insert zero or an invented cap.
|
|
187
|
+
|
|
188
|
+
Discover actual entitled structure keys and supported controls first:
|
|
189
|
+
|
|
190
|
+
```bash
|
|
191
|
+
exodus video start --conceit ugc-template --input-options --json > options.json
|
|
192
|
+
STRUCTURE="$(jq -r '.templateExperience.defaultStructureKey' options.json)"
|
|
193
|
+
```
|
|
194
|
+
|
|
195
|
+
Review that structure and its locks. Then quote a new brief (the original text
|
|
196
|
+
remains `brief.txt` throughout):
|
|
197
|
+
|
|
198
|
+
```bash
|
|
199
|
+
exodus video start --script brief.txt --conceit ugc-template --style raw-iphone-ugc-template \
|
|
200
|
+
--describe --input-mode hybrid --subject product --structure "$STRUCTURE" \
|
|
201
|
+
--requested-seconds 30 --quote-preparation --json > script-quote.json
|
|
202
|
+
jq '.preparationQuote | {estimatedUsd, maximumUsd, scope, expiresAt}' script-quote.json
|
|
203
|
+
# Only after explicitly reviewing and accepting this phase's cost:
|
|
204
|
+
exodus video start --script brief.txt --conceit ugc-template --style raw-iphone-ugc-template \
|
|
205
|
+
--describe --input-mode hybrid --subject product --structure "$STRUCTURE" \
|
|
206
|
+
--requested-seconds 30 --authorize-preparation script-quote.json --json > script-review.json
|
|
207
|
+
jq -j '.script' script-review.json > reviewed.txt
|
|
208
|
+
```
|
|
209
|
+
|
|
210
|
+
For paste adaptation, replace `--describe` with `--adapt "<what may change>"` in
|
|
211
|
+
**both** commands and point `--script` at the unchanged original. Compare
|
|
212
|
+
`.templateIntent.input.originalText` with `.script` before continuing. Keep the
|
|
213
|
+
original and resulting files; for a full textual diff, `diff -u original.txt
|
|
214
|
+
reviewed.txt` shows their actual bytes/lines rather than an invented summary.
|
|
215
|
+
To intentionally retain the original, include `--keep-original` in both commands.
|
|
216
|
+
|
|
217
|
+
If you edit the result or change context, first quote and authorize a fresh guarded
|
|
218
|
+
script review, retaining the original source through `--resume-script`:
|
|
219
|
+
|
|
220
|
+
```bash
|
|
221
|
+
exodus video start --script reviewed.txt --resume-script script-review.json \
|
|
222
|
+
--conceit ugc-template --style raw-iphone-ugc-template --stop-at-script \
|
|
223
|
+
--quote-preparation --json > edit-quote.json
|
|
224
|
+
# Review edit-quote.json before accepting its bound:
|
|
225
|
+
exodus video start --script reviewed.txt --resume-script script-review.json \
|
|
226
|
+
--conceit ugc-template --style raw-iphone-ugc-template --stop-at-script \
|
|
227
|
+
--authorize-preparation edit-quote.json --json > refreshed-review.json
|
|
228
|
+
jq -j '.script' refreshed-review.json > reviewed.txt
|
|
229
|
+
```
|
|
230
|
+
|
|
231
|
+
Use the latest reviewed response and its exact script. A script-phase quote is
|
|
232
|
+
**never** storyboard authority. Quote storyboard preparation separately; only
|
|
233
|
+
after its actual scope/cap is available and accepted can this prepare a storyboard
|
|
234
|
+
(and supported prerequisite audio), still stopping for storyboard review:
|
|
235
|
+
|
|
236
|
+
```bash
|
|
237
|
+
exodus video start --script reviewed.txt --resume-script refreshed-review.json \
|
|
238
|
+
--conceit ugc-template --style raw-iphone-ugc-template \
|
|
239
|
+
--quote-preparation --json > storyboard-quote.json
|
|
240
|
+
# Review storyboard-quote.json; do not continue if the server returned a hold.
|
|
241
|
+
exodus video start --script reviewed.txt --resume-script refreshed-review.json \
|
|
242
|
+
--conceit ugc-template --style raw-iphone-ugc-template \
|
|
243
|
+
--authorize-preparation storyboard-quote.json
|
|
244
|
+
```
|
|
245
|
+
|
|
246
|
+
`--rewrite-to-fit` on V2 requires a signed saved review and itself stops for
|
|
247
|
+
review; it cannot launch or do an initial tidy-then-rewrite under one quote.
|
|
248
|
+
When supported, `--source-record` remains the original source-record JSON; V2
|
|
249
|
+
retains it through script stops and includes it in the storyboard quote. It is
|
|
250
|
+
not silently made adaptation evidence or production rights. The server judges
|
|
251
|
+
whether the selected format supports it. Ordinary Door 1/Swipe/Show commands keep
|
|
252
|
+
their existing behavior.
|
|
253
|
+
|
|
254
|
+
#### Existing template inputs and continuation
|
|
255
|
+
|
|
256
|
+
`video start --script <file> --conceit <template> --style <look>` accepts
|
|
257
|
+
`--requested-seconds <seconds>`, `--pictures-per-scene <1|3>` (default 1), and
|
|
258
|
+
`--product <brandProductId>` for template runs. Requested seconds are a positive
|
|
259
|
+
finite number, including fractions: the CLI does not clamp them to duration
|
|
260
|
+
presets. The server can refuse a length the selected template cannot support.
|
|
261
|
+
`--duration` remains the finished-cut upload duration, not a template input.
|
|
262
|
+
|
|
263
|
+
Choose an existing product belonging to your brand, not a product name or image
|
|
264
|
+
URL. Its authorized product photos are the supported product-reference path;
|
|
265
|
+
they are not actor identities, style references, or image-ad inputs. These flags
|
|
266
|
+
do not change ordinary script/swipe starts or Show runs and are refused on
|
|
267
|
+
`--show`. Missing values, invalid counts/numbers, and conflicting repeated choices
|
|
268
|
+
are refused before networking.
|
|
269
|
+
|
|
270
|
+
All metadata-identified approved templates require a free phase quote and explicit
|
|
271
|
+
authorization of that saved quote before paid preparation, even when using none
|
|
272
|
+
of the new input-mode flags. Cost-consent flags do not select an input mode,
|
|
273
|
+
invent `templateIntent`, or turn an existing script into a writing brief.
|
|
274
|
+
For a new start without `--describe` or `--adapt`, the supplied script remains
|
|
275
|
+
verbatim. A saved receipt retains its exact original source and source mode.
|
|
276
|
+
The first authorized script check always stops for review; it does not launch
|
|
277
|
+
a storyboard. `--one-shot` is refused, and `--rewrite-to-fit` needs a signed
|
|
278
|
+
saved review rather than an initial tidy-and-rewrite under one quote.
|
|
279
|
+
Authenticated template discovery is required; a failed catalog read is not a
|
|
280
|
+
paid-path bypass.
|
|
281
|
+
|
|
282
|
+
`--stop-at-script` starts no storyboard or video run, but writing and claim
|
|
283
|
+
checking may still incur the quoted provider costs. Save the entire JSON
|
|
284
|
+
response unchanged: it retains `scriptApproval`, exact original
|
|
285
|
+
`approvedSource`, `templateBrief` and `productId` when supplied. Use `jq -j` to
|
|
286
|
+
copy the exact script without adding a newline:
|
|
287
|
+
|
|
288
|
+
```bash
|
|
289
|
+
exodus video start --script script.txt --conceit ugc-template --style raw-iphone-ugc-template \
|
|
290
|
+
--requested-seconds 45 --pictures-per-scene 1 --product <brandProductId> \
|
|
291
|
+
--stop-at-script --quote-preparation --json > script-quote.json
|
|
292
|
+
# Inspect the phase, maximumUsd and exact input before explicitly authorizing.
|
|
293
|
+
exodus video start --script script.txt --conceit ugc-template --style raw-iphone-ugc-template \
|
|
294
|
+
--requested-seconds 45 --pictures-per-scene 1 --product <brandProductId> \
|
|
295
|
+
--stop-at-script --authorize-preparation script-quote.json --json > saved-stop-response.json
|
|
296
|
+
jq -j '.script' saved-stop-response.json > reviewed.txt
|
|
297
|
+
# Read reviewed.txt. An unchanged approved script needs a separate storyboard quote.
|
|
298
|
+
exodus video start --script reviewed.txt --resume-script saved-stop-response.json \
|
|
299
|
+
--conceit ugc-template --style raw-iphone-ugc-template \
|
|
300
|
+
--quote-preparation --json > storyboard-quote.json
|
|
301
|
+
exodus video start --script reviewed.txt --resume-script saved-stop-response.json \
|
|
302
|
+
--conceit ugc-template --style raw-iphone-ugc-template \
|
|
303
|
+
--authorize-preparation storyboard-quote.json
|
|
304
|
+
```
|
|
305
|
+
|
|
306
|
+
If you edit the script or its bound context, first quote and authorize another
|
|
307
|
+
script check using `--resume-script` with **`--stop-at-script --json`**. Review and
|
|
308
|
+
save that refreshed response before separately quoting storyboard preparation.
|
|
309
|
+
Neither a quote nor a previously signed receipt silently reapproves changed copy.
|
|
310
|
+
The resumed POST carries the script alongside the saved opaque `scriptApproval`,
|
|
311
|
+
exact approved source, brief and product. The server checks its workspace,
|
|
312
|
+
template and context binding. An approved description stays the original
|
|
313
|
+
description, not the generated or edited script. Source modes are `verbatim` and
|
|
314
|
+
`description`; `script` is not a
|
|
315
|
+
wire mode. Explicit `--requested-seconds`, `--pictures-per-scene`, `--voice`,
|
|
316
|
+
`--voice-path` and `--video-model` override their matching saved choices. The CLI
|
|
317
|
+
retains the selected video model returned in the stopped response alongside
|
|
318
|
+
voice mode/path; a response without a model leaves the server to choose its
|
|
319
|
+
default. Malformed saved model values are refused locally. The server still
|
|
320
|
+
checks supported models and compatibility with the selected voice route.
|
|
321
|
+
These route choices are separate from signed account-voice selections.
|
|
322
|
+
Changing `--product`
|
|
323
|
+
requires a fresh write and approval without `--resume-script`; an old approval
|
|
324
|
+
cannot be relabelled. Outside V2, repeat other desired launch flags and supply
|
|
325
|
+
original `--source-record` / `--adaptation-plan` files when needed. V2 retains these
|
|
326
|
+
declared context records, but never unrelated saved execution/trust fields.
|
|
327
|
+
|
|
328
|
+
With authenticated template metadata, `--style` is also included in the writer's
|
|
329
|
+
template brief when it differs from that template's default. For legacy no-intent
|
|
330
|
+
starts, returning to the current server-provided default removes the saved style
|
|
331
|
+
override. V2 retains an already explicit brief-style pin for server validation
|
|
332
|
+
and material-context review. The server still validates style compatibility and
|
|
333
|
+
template locks; the CLI keeps no local catalog or hardcoded default.
|
|
334
|
+
|
|
335
|
+
Templates also accept existing saved references and account voices:
|
|
336
|
+
|
|
337
|
+
List supported saved references and actual account voices without writing or
|
|
338
|
+
starting anything:
|
|
339
|
+
|
|
340
|
+
```bash
|
|
341
|
+
exodus video start --conceit ugc-template --input-options --voice voice-first --json
|
|
342
|
+
```
|
|
343
|
+
|
|
344
|
+
This read-only command needs no script or Look. Optional `--voice`, `--voice-path`
|
|
345
|
+
and `--video-model` select the route whose capabilities are listed. Unsupported
|
|
346
|
+
selectors, an unavailable voice account and an available-but-empty catalog remain
|
|
347
|
+
distinct; no substitute voice catalog is supplied. Use the returned IDs below.
|
|
348
|
+
|
|
349
|
+
```bash
|
|
350
|
+
exodus video start --script script.txt --conceit ugc-template --style raw-iphone-ugc-template \
|
|
351
|
+
--elements '[{"handle":"maya","elementId":"<savedElementId>"}]' \
|
|
352
|
+
--voices '{"MAYA":"<accountVoiceId>"}' --stop-at-script --quote-preparation --json > selected-quote.json
|
|
353
|
+
# Inspect selected-quote.json before explicitly authorizing the same input.
|
|
354
|
+
exodus video start --script script.txt --conceit ugc-template --style raw-iphone-ugc-template \
|
|
355
|
+
--elements '[{"handle":"maya","elementId":"<savedElementId>"}]' \
|
|
356
|
+
--voices '{"MAYA":"<accountVoiceId>"}' --stop-at-script \
|
|
357
|
+
--authorize-preparation selected-quote.json --json > selected-stop.json
|
|
358
|
+
```
|
|
359
|
+
|
|
360
|
+
Both flags accept inline JSON or `@file.json`. References use authorized saved
|
|
361
|
+
character/location IDs, not image uploads. Voice-map keys are canonical script
|
|
362
|
+
speaker names; `NARRATOR` means offscreen narration, not a visible character.
|
|
363
|
+
A string pins a voice ID, `{ "voiceId": "...", "label": "..." }` adds a label,
|
|
364
|
+
and `null` clears that speaker's pin. The server validates account availability
|
|
365
|
+
and template support; the CLI does not invent voice IDs or a catalog endpoint.
|
|
366
|
+
|
|
367
|
+
Saved references and account-voice maps survive edited-script continuation and
|
|
368
|
+
are bound to the approval. To change or remove a selection, resume with the new
|
|
369
|
+
`--elements` / `--voices` and **`--stop-at-script --json`**, first quoting and then
|
|
370
|
+
explicitly authorizing that script-check phase. This checks the edited draft
|
|
371
|
+
against its original source and new selections and returns a new bound receipt.
|
|
372
|
+
Save that response before separately quoting storyboard preparation. A direct start with
|
|
373
|
+
changed selections refuses before networking; it never silently reapproves or
|
|
374
|
+
discards the edited file. Equivalent explicit choices can repeat saved choices.
|
|
375
|
+
Repeated selection flags use the last value; `[]` references and `{}` voices
|
|
376
|
+
mean no selection and are omitted from the request. Changing the product still
|
|
377
|
+
requires a separate fresh write without `--resume-script`.
|
|
378
|
+
|
|
379
|
+
Resume requires `--script`, `--conceit` and `--style`, and always checks the
|
|
380
|
+
authenticated template menu before posting. A saved `launchDoor` or trust stamp
|
|
381
|
+
does not authorize a launch. Invalid JSON, malformed saved source/brief/product,
|
|
382
|
+
missing receipt, conflicting file flags, changed product overrides and Show
|
|
383
|
+
resumes are refused before networking. The saved response must have
|
|
384
|
+
`stoppedAtScript: true`, nonempty `script` and `scriptApproval` strings, and an
|
|
385
|
+
`approvedSource` with a valid mode and nonempty text. Older responses without a
|
|
386
|
+
receipt require a fresh write/approval. A brief may contain only string-valued
|
|
387
|
+
`cards` and primitive-valued `settings`; the server still judges the choices and
|
|
388
|
+
verifies the opaque receipt. Do not edit or fabricate it.
|
|
389
|
+
|
|
43
390
|
### Listen to and approve a run's Song
|
|
44
391
|
|
|
45
392
|
A run that sings its ad holds one Song master. Listen before motion starts:
|
|
46
393
|
|
|
47
394
|
- `video song <runId>` prints the Song's sha256s, its lyric, which lyric lines
|
|
48
395
|
were heard, and whether it can be approved. It never downloads the audio.
|
|
396
|
+
If recovery stopped before a master exists, it prints the held reason instead
|
|
397
|
+
(`song: null` and a separate `hold` object in JSON, safe for older CLIs).
|
|
398
|
+
Use **Try again on the held voiceover step**
|
|
399
|
+
to resume this run's paid task and retained tracks; do not approve the empty
|
|
400
|
+
checkpoint or start a new run to recover it. A retry does not replace a song;
|
|
401
|
+
an unchanged permanent hold needs its stated input or recovery action first.
|
|
402
|
+
Section alternatives longer than the approved transcription clock are retained
|
|
403
|
+
but skipped without transcription. A fitting alternative can still become the
|
|
404
|
+
master. If none fits, the hold lists both decoded lengths and the allowed clock;
|
|
405
|
+
retry rechecks the retained alternatives without ordering another song.
|
|
49
406
|
- `video song approve <runId> --master <masterSha256>` approves exactly the
|
|
50
407
|
master you listened to. If the run now holds a different one, nothing is
|
|
51
408
|
approved.
|