@tanstack/ai-byteplus 0.0.0

package/src/video/video-provider-options.ts

@@ -0,0 +1,361 @@
+/**
+ * Provider options and per-model capability tables for the BytePlus Seedance
+ * video models.
+ *
+ * Every applicability claim below was probed live against
+ * `https://ark.ap-southeast.bytepluses.com/api/v3` on 2026-07-31. The probe
+ * sent an out-of-range `seed` alongside the field under test, so requests that
+ * passed validation still failed before a task was created (nothing billed):
+ * an error naming the field under test means "rejected", an error naming
+ * `seed` means "accepted". Ark reports only one arbitrary invalid parameter
+ * per request, so each cell was retried until a verdict repeated.
+ *
+ * Ark rejects an inapplicable field outright — "the specified parameter
+ * `draft` is not supported for model seedance-1-0-pro in t2v, must be empty" —
+ * so these tables are not cosmetic: sending a field to the wrong model is a
+ * 400, not a no-op.
+ *
+ * **Where the adapter guards, and where it doesn't** (deliberate, not an
+ * oversight). Scalar applicability — `service_tier`, `draft`, `priority`,
+ * `frames`, `camera_fixed` — is left to Ark, whose 400 names the offending
+ * field and the model precisely enough to act on, and whose per-model rules
+ * shift as BytePlus ships models. Duplicating that here would mean a table
+ * that silently goes stale and starts rejecting requests the API would have
+ * accepted. The adapter guards locally only where the API's own error is
+ * misleading or arrives too late to be actionable: prompt media shape (role
+ * vocabulary, frame-vs-reference exclusivity, frame cardinality) and the
+ * resolution tier, both of which are derived from a caller's `prompt` /
+ * `size` rather than passed through verbatim.
+ *
+ * @experimental Video generation is an experimental feature and may change.
+ */
+
+import { isKnownBytePlusVideoModel } from '../model-meta'
+import type {
+  BytePlusVideoModel,
+  BytePlusVideoModelOrString,
+  BytePlusVideoRatio,
+  BytePlusVideoResolution,
+} from '../model-meta'
+
+/**
+ * Inference queue for the request.
+ *
+ * - `default` — online inference: lower RPM and concurrency quotas, lowest
+ *   latency.
+ * - `flex` — offline batch inference: higher daily token quotas at half the
+ *   price, with no latency guarantee. Task ids come back with a `cgt-batch-`
+ *   prefix (live-verified).
+ *
+ * Only the Seedance 1.x models accept this field. The Seedance 2.0 family
+ * rejects it ("service_tier is not supported … must be empty").
+ *
+ * @experimental Video generation is an experimental feature and may change.
+ */
+export type BytePlusVideoServiceTier = 'default' | 'flex'
+
+/**
+ * Provider-specific options for Seedance video generation. These map one-to-one
+ * onto the create-task request body and take precedence over the values the
+ * adapter derives from the generic `size` / `duration` options.
+ *
+ * Fields are model-dependent; each one documents where it applies. Passing a
+ * field to a model that does not accept it is a 400 from Ark, not a silent
+ * ignore.
+ *
+ * @experimental Video generation is an experimental feature and may change.
+ */
+export interface BytePlusVideoProviderOptions {
+  /**
+   * Output aspect ratio. Overrides the ratio half of the generic `size`.
+   *
+   * `adaptive` (follow the input frame) is the default on Seedance 2.0 and
+   * 1.5-pro but is rejected by Seedance 1.0-pro / 1.0-pro-fast for
+   * text-to-video.
+   */
+  ratio?: BytePlusVideoRatio
+
+  /**
+   * Output resolution tier. Overrides the resolution half of the generic
+   * `size`. Matched case-insensitively by the API; this package uses
+   * lowercase throughout. `4k` exists only on `dreamina-seedance-2-0-260128`,
+   * and there is no 2K tier on any model.
+   */
+  resolution?: BytePlusVideoResolution
+
+  /**
+   * Whole seconds of output. Overrides the generic `duration`, and unlike it
+   * is sent verbatim rather than snapped into the model's range.
+   *
+   * `-1` asks the model to choose its own length; accepted by Seedance 2.0
+   * and 1.5-pro only.
+   */
+  duration?: number
+
+  /**
+   * Frame count instead of `duration`, for fractional-second output. Takes
+   * precedence over `duration` server-side. Valid values are the integers of
+   * the form `25 + 4n` within `[29, 289]`, at 24 fps.
+   *
+   * Seedance 1.0-pro and 1.0-pro-fast only.
+   */
+  frames?: number
+
+  /**
+   * Randomness seed, an integer in `[-1, 2^32-1]`. `-1` (the default) leaves
+   * generation unseeded. Accepted by every Seedance model.
+   */
+  seed?: number
+
+  /**
+   * Appends a "fix the camera" instruction to the prompt. Best-effort — the
+   * model is not constrained to obey it.
+   *
+   * Seedance 1.5-pro, 1.0-pro and 1.0-pro-fast only; the 2.0 family rejects
+   * it.
+   */
+  camera_fixed?: boolean
+
+  /** Burn a watermark into the output. Defaults to `false`. */
+  watermark?: boolean
+
+  /**
+   * Generate an audio track synchronized with the visuals — dialogue, effects
+   * and score inferred from the prompt. Quote dialogue in the prompt for
+   * better results.
+   *
+   * Accepted by every model at the API's validation layer, but only Seedance
+   * 2.0 and 1.5-pro actually produce audio.
+   */
+  generate_audio?: boolean
+
+  /**
+   * Inference queue. Seedance 1.x only — the 2.0 family has no offline tier.
+   */
+  service_tier?: BytePlusVideoServiceTier
+
+  /**
+   * Also return the video's final frame as a watermark-free PNG, readable from
+   * the finished task as `content.last_frame_url`. Chain it into the next
+   * task's first frame to extend a shot. Accepted by every model.
+   */
+  return_last_frame?: boolean
+
+  /**
+   * Render a cheap, low-fidelity preview to sanity-check staging and camera
+   * work before paying for the real thing.
+   *
+   * Seedance 1.5-pro only.
+   */
+  draft?: boolean
+
+  /**
+   * Queue priority, `[0, 9]`. Seedance 2.0 family only — 1.5-pro rejects it,
+   * and the 1.0 models accept it without acting on it.
+   */
+  priority?: number
+
+  /**
+   * Seconds after `created_at` at which an unfinished task is abandoned and
+   * marked `expired`. Documented range `[3600, 259200]`, default 172800
+   * (48 hours). The floor is enforced on Seedance 1.x but not on the 2.0
+   * family.
+   */
+  execution_expires_after?: number
+
+  /**
+   * URL that receives a POST with the full task payload on every status
+   * change. BytePlus retries a failed delivery three times.
+   */
+  callback_url?: string
+
+  /**
+   * Stable opaque identifier for the end user driving the request, for abuse
+   * attribution. Max 64 characters — hash the real identifier rather than
+   * sending it.
+   */
+  safety_identifier?: string
+}
+
+/**
+ * Type-only map from video model name to its provider options. Seedance takes
+ * the same option surface across models; applicability is per-field and
+ * documented on {@link BytePlusVideoProviderOptions}.
+ *
+ * @experimental Video generation is an experimental feature and may change.
+ */
+export type BytePlusVideoModelProviderOptionsByName = {
+  [K in BytePlusVideoModel]: BytePlusVideoProviderOptions
+}
+
+/**
+ * Aspect ratios accepted by the create endpoint.
+ *
+ * `adaptive` is rejected by Seedance 1.0-pro / 1.0-pro-fast for text-to-video
+ * but is the documented default for their image-to-video path, so it is not
+ * filtered per model here.
+ */
+const BYTEPLUS_VIDEO_RATIOS: ReadonlyArray<string> = [
+  '16:9',
+  '9:16',
+  '4:3',
+  '3:4',
+  '1:1',
+  '21:9',
+  'adaptive',
+]
+
+/**
+ * Resolutions each model accepts, live-probed.
+ *
+ * Two findings here contradict the BytePlus prose docs and are worth calling
+ * out: there is no 2K tier on any Seedance model (`2k`/`2K` is rejected
+ * everywhere, including on the 2.0 flagship whose docs advertise "up to 4K"),
+ * and `seedance-1-0-pro-fast-251015` does accept `1080p` despite being
+ * documented as 480p/720p only.
+ */
+const BYTEPLUS_VIDEO_RESOLUTIONS: {
+  readonly [K in BytePlusVideoModel]: ReadonlyArray<BytePlusVideoResolution>
+} = {
+  'dreamina-seedance-2-0-260128': ['480p', '720p', '1080p', '4k'],
+  'dreamina-seedance-2-0-fast-260128': ['480p', '720p'],
+  'dreamina-seedance-2-0-mini-260615': ['480p', '720p'],
+  'seedance-1-5-pro-251215': ['480p', '720p', '1080p'],
+  'seedance-1-0-pro-250528': ['480p', '720p', '1080p'],
+  'seedance-1-0-pro-fast-251015': ['480p', '720p', '1080p'],
+}
+
+/**
+ * Models accepting reference-media mode (`r2v`): reference images, video and
+ * audio that the output draws on without pinning specific frames. The 1.x
+ * models reject it with "the specified task_type r2v does not support model …".
+ */
+const BYTEPLUS_VIDEO_REFERENCE_MEDIA_MODELS: ReadonlySet<string> = new Set([
+  'dreamina-seedance-2-0-260128',
+  'dreamina-seedance-2-0-fast-260128',
+  'dreamina-seedance-2-0-mini-260615',
+])
+
+/**
+ * Models accepting a closing frame (`flf2v`, first-and-last-frame mode).
+ * `seedance-1-0-pro-fast-251015` is the one Seedance model without it — it
+ * does text-to-video and single-first-frame image-to-video only.
+ */
+const BYTEPLUS_VIDEO_LAST_FRAME_MODELS: ReadonlySet<string> = new Set([
+  'dreamina-seedance-2-0-260128',
+  'dreamina-seedance-2-0-fast-260128',
+  'dreamina-seedance-2-0-mini-260615',
+  'seedance-1-5-pro-251215',
+  'seedance-1-0-pro-250528',
+])
+
+/**
+ * True when the model is *known* to support reference-media mode (reference
+ * images, video and audio). An id this package has no metadata for answers
+ * `false`; callers must decide whether that means "no" or "unknown" — the
+ * adapter treats it as unknown and lets Ark rule.
+ *
+ * @experimental Video generation is an experimental feature and may change.
+ */
+export function supportsReferenceMedia(model: string): boolean {
+  return BYTEPLUS_VIDEO_REFERENCE_MEDIA_MODELS.has(model)
+}
+
+/**
+ * True when the model is *known* to support pinning the video's closing
+ * frame. Same unknown-id caveat as {@link supportsReferenceMedia}.
+ *
+ * @experimental Video generation is an experimental feature and may change.
+ */
+export function supportsLastFrame(model: string): boolean {
+  return BYTEPLUS_VIDEO_LAST_FRAME_MODELS.has(model)
+}
+
+/**
+ * Splits a `size` template into its Seedance request fields.
+ *
+ * The template is either a bare aspect ratio (`'16:9'`) or
+ * `ratio_resolution` (`'16:9_720p'`), mirroring the grok video adapter.
+ * Returns `undefined` when the string doesn't match the template at all.
+ *
+ * The resolution half comes back lowercased. Ark itself matches the field
+ * case-insensitively, but this package standardizes on lowercase so callers
+ * can compare the result against {@link BytePlusVideoResolution} directly.
+ *
+ * @experimental Video generation is an experimental feature and may change.
+ */
+export function parseBytePlusVideoSize(
+  size: string,
+): { ratio: string; resolution?: string } | undefined {
+  const match = /^(\d+:\d+|adaptive)(?:_(.+))?$/.exec(size)
+  const [, ratio, resolution] = match ?? []
+  if (ratio === undefined) return undefined
+  return {
+    ratio,
+    ...(resolution !== undefined && { resolution: resolution.toLowerCase() }),
+  }
+}
+
+/**
+ * Validates a resolution against a model's tiers, returning it lowercased.
+ *
+ * Used for both halves of the request: the resolution parsed out of the
+ * generic `size`, and a `modelOptions.resolution` that overrides it. A model
+ * this package has no table for is normalized but not checked — see
+ * {@link BytePlusVideoModelOrString}.
+ *
+ * @throws Error when a known model does not offer the tier.
+ *
+ * @experimental Video generation is an experimental feature and may change.
+ */
+export function resolveBytePlusVideoResolution(
+  model: BytePlusVideoModelOrString,
+  resolution: string,
+): string {
+  const normalized = resolution.toLowerCase()
+  if (!isKnownBytePlusVideoModel(model)) return normalized
+
+  const allowed = BYTEPLUS_VIDEO_RESOLUTIONS[model]
+  if (!allowed.includes(normalized as BytePlusVideoResolution)) {
+    throw new Error(
+      `byteplus: resolution "${resolution}" is not supported by model ` +
+        `"${model}". Supported resolutions: ${allowed.join(', ')}.`,
+    )
+  }
+  return normalized
+}
+
+/**
+ * Validates a `size` template against a model and returns the request fields
+ * it maps onto, with the resolution lowercased.
+ *
+ * For an unknown model only the template's *shape* is checked — enough to
+ * split it into `ratio` and `resolution` — because a future model may bring
+ * ratios and tiers that do not exist today. Ark validates the values.
+ *
+ * @throws Error when the template is malformed, or (known models only) the
+ * ratio is unknown or the resolution is not offered by this model.
+ *
+ * @experimental Video generation is an experimental feature and may change.
+ */
+export function resolveBytePlusVideoSize(
+  model: BytePlusVideoModelOrString,
+  size: string,
+): { ratio: string; resolution?: string } {
+  const parsed = parseBytePlusVideoSize(size)
+  const known = isKnownBytePlusVideoModel(model)
+  if (!parsed || (known && !BYTEPLUS_VIDEO_RATIOS.includes(parsed.ratio))) {
+    throw new Error(
+      `byteplus: size "${size}" is not supported by model "${model}". Expected ` +
+        `"ratio" or "ratio_resolution" (e.g. "16:9_720p") with ratio one of: ` +
+        `${BYTEPLUS_VIDEO_RATIOS.join(', ')}.`,
+    )
+  }
+
+  return {
+    ratio: parsed.ratio,
+    ...(parsed.resolution !== undefined && {
+      resolution: resolveBytePlusVideoResolution(model, parsed.resolution),
+    }),
+  }
+}

package/src/video/wire-types.ts

@@ -0,0 +1,293 @@
+/**
+ * Wire types for the BytePlus Ark Seedance video task API
+ * (`/contents/generations/tasks`).
+ *
+ * Hand-written minimal shapes covering only the fields this adapter sends and
+ * reads, with provenance noted inline. Three sources:
+ *
+ * 1. The harvested OpenAPI 3.1 documents for the `ark` service, actions
+ *    `CreateContentsGenerationsTasks` (`x-updated-time: 2026-05-07`),
+ *    `GetContentsGenerationsTask` (`2026-04-14`),
+ *    `ListContentsGenerationsTasks` (`2026-03-24`) and
+ *    `DeleteContentsGenerationsTasks` — authoritative for field names,
+ *    defaults and response shapes.
+ * 2. Live calls against `https://ark.ap-southeast.bytepluses.com/api/v3` on
+ *    2026-07-31 with a real `ARK_API_KEY`, which pinned the create response,
+ *    the per-model parameter applicability (see
+ *    `video-provider-options.ts`) and the `content[]` role vocabulary.
+ * 3. The Seedance prose docs, for the retention windows.
+ *
+ * Two casing traps worth knowing: the response frame-rate field is
+ * `framespersecond` (all lowercase, no underscores), and `resolution` is
+ * matched case-insensitively on the way in (`4K`, `4k` and even `1080P` are
+ * all accepted — live-verified), so this package standardizes on lowercase.
+ */
+
+/**
+ * Task lifecycle states.
+ *
+ * `queued` and `running` are non-terminal; the rest are terminal. `cancelled`
+ * records are dropped 24 hours after cancellation, and only a `queued` task
+ * can be cancelled at all.
+ *
+ * Source: `GetContentsGenerationsTask` / `ListContentsGenerationsTasks`
+ * `status` descriptions. (The Get document omits `expired` from its list
+ * while the List document includes it; `execution_expires_after` is documented
+ * as producing `expired` on both, so it is included here.)
+ */
+export type BytePlusVideoTaskStatus =
+  | 'queued'
+  | 'running'
+  | 'succeeded'
+  | 'failed'
+  | 'cancelled'
+  | 'expired'
+
+/**
+ * Role of a media item inside `content[]`.
+ *
+ * Live-verified: an unknown role is rejected with "invalid role specified for
+ * image content", and the API sorts requests into task types from the roles
+ * present — `i2v` (first frame), `flf2v` (first + last frame) and `r2v`
+ * (reference media). The two families are mutually exclusive: mixing them
+ * fails with "first/last frame content cannot be mixed with reference media
+ * content".
+ */
+export type BytePlusVideoContentRole =
+  | 'first_frame'
+  | 'last_frame'
+  | 'reference_image'
+  | 'reference_video'
+  | 'reference_audio'
+
+/** Instruction text for the generation. */
+export interface BytePlusVideoTextContent {
+  type: 'text'
+  text: string
+}
+
+/**
+ * An image input. A public URL is fetched by BytePlus server-side; a
+ * `data:` URI carries the bytes inline.
+ */
+export interface BytePlusVideoImageContent {
+  type: 'image_url'
+  image_url: { url: string }
+  /** Omitted for a bare first frame — the API defaults to `first_frame`. */
+  role?: BytePlusVideoContentRole
+}
+
+/** A video input. Reference-media mode requires `role: 'reference_video'`. */
+export interface BytePlusVideoVideoContent {
+  type: 'video_url'
+  video_url: { url: string }
+  role?: BytePlusVideoContentRole
+}
+
+/**
+ * An audio input. Live-verified: audio can only accompany another reference
+ * input — "reference_audio cannot be the only reference input".
+ */
+export interface BytePlusVideoAudioContent {
+  type: 'audio_url'
+  audio_url: { url: string }
+  role?: BytePlusVideoContentRole
+}
+
+/**
+ * One entry of the `content[]` array.
+ *
+ * The create schema declares `maxItems: 5`, but the live API does not enforce
+ * it — 7 entries (6 reference images plus text) were accepted on
+ * `dreamina-seedance-2-0-260128`. The adapter therefore does not cap the array
+ * locally; a genuinely over-long request gets whatever Ark decides to say.
+ */
+export type BytePlusVideoContentPart =
+  | BytePlusVideoTextContent
+  | BytePlusVideoImageContent
+  | BytePlusVideoVideoContent
+  | BytePlusVideoAudioContent
+
+/**
+ * Request body for `POST /contents/generations/tasks`.
+ *
+ * Only `model` and `content` are required. Every other field is
+ * model-dependent — Ark rejects an inapplicable field outright ("the
+ * specified parameter `draft` is not supported for model … must be empty")
+ * rather than ignoring it, so the adapter only sends what the caller asked
+ * for. See `video-provider-options.ts` for the live-probed applicability
+ * matrix.
+ */
+export interface BytePlusVideoCreateRequest {
+  /** Seedance model id (or a preconfigured endpoint id). */
+  model: string
+
+  /** Prompt text plus any image / video / audio inputs, max 5 entries. */
+  content: Array<BytePlusVideoContentPart>
+
+  /** Output aspect ratio, e.g. `16:9`. `adaptive` follows the input frame. */
+  ratio?: string
+
+  /** Resolution tier, e.g. `720p`. Matched case-insensitively by the API. */
+  resolution?: string
+
+  /** Whole seconds of output. `-1` lets the model choose (Seedance 2.0 / 1.5). */
+  duration?: number
+
+  /** Frame count, an alternative to `duration` that allows fractional seconds. */
+  frames?: number
+
+  /** Randomness seed; integers in `[-1, 2^32-1]`, where `-1` means unseeded. */
+  seed?: number
+
+  /** Appends a "fix the camera" instruction to the prompt. Default `false`. */
+  camera_fixed?: boolean
+
+  /** Burn a watermark into the output. Default `false`. */
+  watermark?: boolean
+
+  /** Generate a synchronized audio track. Default `false`. */
+  generate_audio?: boolean
+
+  /** `default` (online) or `flex` (offline batch, half price). */
+  service_tier?: string
+
+  /** Also return the final frame as a PNG. Default `false`. */
+  return_last_frame?: boolean
+
+  /** Cheap low-fidelity preview render. Default `false`. */
+  draft?: boolean
+
+  /** Queue priority `[0, 9]`. */
+  priority?: number
+
+  /** Seconds from `created_at` after which the task is marked `expired`. */
+  execution_expires_after?: number
+
+  /** URL that receives a POST with the task payload on each status change. */
+  callback_url?: string
+
+  /** Opaque per-end-user identifier for abuse attribution, max 64 chars. */
+  safety_identifier?: string
+}
+
+/**
+ * Response of `POST /contents/generations/tasks`.
+ *
+ * Live-verified: the body is just the task id (e.g.
+ * `cgt-batch-20260731174311-zmz5s`; the `-batch` infix appears when the task
+ * is routed to the `flex` offline queue).
+ */
+export interface BytePlusVideoCreateResponse {
+  id?: string
+}
+
+/**
+ * Error detail attached to a terminal task.
+ *
+ * Codes are the dotted/PascalCase Ark strings — the create, get and list
+ * documents enumerate `InputTextSensitiveContentDetected`,
+ * `InputImageSensitiveContentDetected`, `OutputVideoSensitiveContentDetected`
+ * and `QuotaExceeded` under `x-error-code`. The set is open-ended, so this
+ * stays a plain `string`.
+ */
+export interface BytePlusVideoTaskError {
+  code?: string
+  message?: string
+}
+
+/**
+ * Token usage for a finished task. Video generation bills output only, so
+ * `total_tokens` equals `completion_tokens` and there is no prompt count.
+ *
+ * The schema types both counts as `string` while the documented example
+ * response shows bare numbers, so both are accepted and coerced.
+ */
+export interface BytePlusVideoTaskUsage {
+  completion_tokens?: number | string
+  total_tokens?: number | string
+  /** Only present when a provider tool (web search) ran. */
+  tool_usage?: { web_search?: number }
+}
+
+/** Output URLs of a succeeded task. Both links expire 24 hours after success. */
+export interface BytePlusVideoTaskContent {
+  /** MP4 download URL. */
+  video_url?: string
+  /** Final frame as PNG; only when `return_last_frame` was set. */
+  last_frame_url?: string
+}
+
+/**
+ * Response of `GET /contents/generations/tasks/{id}`.
+ *
+ * `content` appears once the task succeeds; `error` appears when it fails.
+ * Note `duration` comes back as a string here while the list endpoint types
+ * it as an integer, so both are accepted.
+ */
+export interface BytePlusVideoTask {
+  id?: string
+  /** `{model name}-{version}` actually used — not necessarily the id sent. */
+  model?: string
+  status?: BytePlusVideoTaskStatus
+  error?: BytePlusVideoTaskError
+  /** Unix seconds. Anchors the 7-day task-record retention. */
+  created_at?: number
+  /** Unix seconds of the last status change — for a succeeded task, when the
+   * output (and its 24-hour URL) was produced. */
+  updated_at?: number
+  content?: BytePlusVideoTaskContent
+  seed?: number
+  resolution?: string
+  ratio?: string
+  duration?: number | string
+  frames?: number
+  /** Frame rate. Lowercase and unseparated on the wire — not `frames_per_second`. */
+  framespersecond?: number
+  generate_audio?: boolean
+  service_tier?: string
+  draft?: boolean
+  draft_task_id?: string
+  execution_expires_after?: number
+  safety_identifier?: string
+  usage?: BytePlusVideoTaskUsage
+
+  // The two fields below came back on a live succeeded task
+  // (`seedance-1-0-pro-fast-251015`, 2026-07-31) but are absent from the
+  // harvested Get schema.
+  /** Queue priority the task ran at. */
+  priority?: number
+  /** Container of the generated video, e.g. `mp4`. */
+  output_format?: string
+}
+
+/**
+ * One entry of the list response.
+ *
+ * The list document declares the same fields as the get document (including
+ * `error` — despite BytePlus prose elsewhere calling the list-side field
+ * `failure_reason`, no such field exists in the harvested schema, so it is
+ * not typed here).
+ */
+export interface BytePlusVideoTaskListItem extends BytePlusVideoTask {}
+
+/**
+ * Response of `GET /contents/generations/tasks`.
+ *
+ * Supported query parameters: `page_num` and `page_size` (both `[1, 500]`),
+ * `filter.status`, repeated `filter.task_ids`, and `filter.service_tier`.
+ *
+ * **`flex` tasks are missing from the default listing.** An unfiltered list
+ * returned `{total: 0, items: []}` both while a live `flex` task was running
+ * and immediately after it succeeded, so offline-tier work is invisible here
+ * unless `filter.service_tier=flex` is passed. Poll a known task id rather
+ * than relying on the listing to discover tasks.
+ */
+export interface BytePlusVideoTaskListResponse {
+  items?: Array<BytePlusVideoTaskListItem>
+  total?: number
+}
+
+// `DELETE /contents/generations/tasks/{id}` cancels a `queued` task and
+// deletes anything already terminal; its documented success body is an empty
+// object, so no response interface is declared for it.