Skip to content

nexu-io/openmotion / skills/openmotion-media

openmotion-media

Verified

Generate, poll and inspect individual image, video, speech or music assets through openmotion_assets_use after openmotion_media_use discovery (configured Atlas Cloud, fal.ai, WaveSpeedAI or OpenDesign Cloud BYOK providers, or a signed-in OpenDesign account), either when delegated by an OpenMotion film workflow or for a standalone asset-generation request. Native-only tasks with no new speech or music stay outside generation unless explicitly requested. Not a whole-film entry point: creating, recreating, templating, editing or exporting a whole video belongs to the OpenMotion video entries.

references/providers/account.md

View on GitHub
# OpenDesign account models

Checked on 2026-11-15 against [OpenDesign source](https://github.com/nexu-io/open-design/tree/59e4f2ae85de7e7318a5e6bdcf3a34584c4b7582) and the command help of `@powerformer/vela-cli@0.1.10`. The CLI owns browser authorization, credential persistence, discovery and most generation submission (image jobs it cannot express are described below); OpenMotion does not reproduce private authentication endpoints or bundle its binary. Account image task reads and resumable result downloads use the service path described below; video and audio task reads remain connector-owned.

The user signs in through **Settings → Model providers → OpenDesign → Sign in with OpenDesign** and completes authorization in their browser. On first use, each workspace prepares the pinned connector in its private data directory (`<data directory>/account/runtime`). Account credentials, session identity and cached models live in the per-user OpenDesign account directory (or `MOTION_ACCOUNT_DIR`, or the workspace's own `account/` under the system temporary directory) so every OpenMotion workspace for the same user shares one sign-in; credentials are not copied from other apps, and are never returned to the Agent. Account models appear under provider ID `account`, alongside BYOK. A successful login establishes identity, not billing or generation acceptance; inspect refreshes billing and returns advisory mediaEntitlement, normalized plan and hasBalance, without raw dollar balances. Free and Go report not-included whatever their balance, because those plans entitle text models only and credits never unlock media models; available stays true for Go text models. unknown (billing failures, an unrecognized plan) continues to discovery. Free or Go users can click **Upgrade to Plus** directly on the same OpenDesign page. Follow [setup guidance](../setup.md) for alternatives.

Call `openmotion_media_use` with `action=inspect`, then `action=reference, provider=account`, then `action=models, provider=account, refresh=true`. Discovery uses the connector's remote model and media catalogues, including their enabled flags. For signed-in accounts, `elevenlabs-v3`, `byteplus-tts-2` and `elevenlabs-music-v2` are also included as built-in (`source: "builtin"`) audio entries whenever the remote media catalogue omits them; a live catalogue entry (including `enabled: false`) takes precedence, and built-in entries do not prove plan entitlement. Use only returned IDs and the provider's model parameter documentation; no model name or entitlement is guessed.

```json
{"action":"generate","requestId":"account-image-one","model":"account:CATALOGUE_MODEL_ID","modality":"image","prompt":"A ceramic cup","parameters":{}}
```

Image parameters supported by the account boundary: `aspect_ratio`, `resolution`, `quality`, `size`, and `background` where the model's `generationOptions` list it (GPT Image 2.6 accepts `"transparent"` for background removal). On the wire, the connector requires `aspect_ratio` and `resolution` together as a pair published for the active operation (`generations` vs `edits`). When the synced catalogue publishes `capabilities.profiles`, OpenMotion validates against the active envelope and automatically completes a missing half from the published default (or first matching) pair before invoking the connector, while rejecting unpublished ratios, resolutions, or pairs before submission. `size` cannot be combined with `aspect_ratio`, `resolution`, or `output`, and `quality` must be one of the model's values; even when no `profiles` are published, OpenMotion rejects incomplete `aspect_ratio`/`resolution` pairs before submission. Omit them to use the model default. Video uses `modality=video`, and accepts `ratio`, `resolution`, `duration`, `generate_audio`. Speech uses `modality=audio`, `mode=text-to-speech`, and `parameters.voice`. For `elevenlabs-v3` and `byteplus-tts-2`, the `models` action lists checked voices and the default voice in `generationOptions`; pick one that suits the language and tone, and omit it only to use the default. Another voice ID belonging to that model is passed through unchecked. Other speech models require a voice ID belonging to that model. Music uses `modality=audio`, `mode=text-to-audio`, and accepts an optional whole-second `parameters.duration`; the catalogue publishes no duration range, so the account service decides which lengths it accepts. Unsupported parameters fail before submission. The account catalogue lists the same video models as OpenDesign Cloud; choose between them as in [Choosing a video model](opendesign.md#choosing-a-video-model).

Image editing and first-frame video: an image model returned with `supportsImageReference: true` (its catalogue entry publishes `edits` for this account) edits input images, up to the `maxInputImages` it reports (the same per-model limits as OpenDesign Cloud, for example one for Seedream 5.5), and a video model returned with `maxInputImages: 1` (its catalogue entry publishes `max_input_images`) accepts one input image as its first frame. Queue either with `openmotion_assets_use` `action=generate`, `model=account:CATALOGUE_MODEL_ID`, and pinned `references` entries (`assetId` and `versionId`); a first-frame video takes exactly one. The image is snapshotted when queued and must be a PNG, JPEG or WebP under 2 MiB; the original asset is not modified. Last-frame and multi-reference video inputs are not supported through the account provider yet.

The service executes one `image gen`, `image edit` (with `--image` set to the snapshotted reference), `video gen` (with `--first-frame` when a first-frame reference is attached), `audio speech` or `audio music` command with `--model`, `--prompt`, `--no-wait`, `--json`. The connector has no background flag and sends one input image, so an image job with `background` or more than one reference is instead uploaded and submitted to the OpenDesign media API (`/api/v1/media/uploads`, `/images/generations` or `/images/edits`) with the same control key, exactly as an OpenDesign Cloud key would. It saves the returned task_id. For images, OpenMotion reads only that original task directly from `/api/v1/media/images/tasks/{id}`, authenticated with the installation connector profile’s control key. It downloads the returned image URL without forwarding credentials, retaining verified partial bytes for bounded retries and the existing `resume` operation; Range/If-Range requires matching asset identity, a verified local checkpoint and a strong ETag. Invalid partial files or validators restart the same image download, never generation. Signed URLs stay in memory and are refreshed through original-task reads. Video and audio still poll and download through the corresponding connector `get` command; account video/audio resume is unavailable. The downloaded file is validated before adding it as a new library asset version; an older project-direct job imports into its original project instead. Poll with the tool that queued the job, as the [job procedure](../jobs.md) describes, and place a selected library result with `openmotion_assets_use action=import_to_project`. Do not submit another task to check progress. Local requestId replay returns the original job.

Account session identity is captured when queuing. Signing out or switching identity prevents an old queued task from submitting under a new account. If `inspect` reports the `account` provider unavailable because the OpenDesign sign-in expired, open the sign-in handoff from [setup](../setup.md) and ask the user to sign in again on the **OpenDesign** provider page. Account tasks queued before that fail without submission; resubmit them after the new sign-in. Failed CLI calls omit private stderr and cannot establish the exact remote error. Check account connectivity, available credits and original task history before retrying. Cancellation stops local work; it does not prove a remote cancellation or refund. Interrupted/uncertain paid requests are never automatically resubmitted or routed to another provider.

For plan or balance failures, follow [account recovery](../jobs.md#opendesign-account-recovery).
planning-and-task-breakdownaddyosmani/agent-skillsBreaks work into ordered tasks. Use when you have a spec or clear requirements and need to break work into implementable tasks. Use when a task feels too large to start, when you need to estimate scope, or when parallel work is possible.Stars: 119kUpdated Claude CodeCodexVerifiedInstall
screenpipethedotmack/claude-memSearch screen and audio history recorded by Screenpipe and optionally save selected context into claude-mem. Use when the user asks to recall desktop activity, a meeting, or something seen in another app through Screenpipe.Stars: 118kUpdated Claude CodeCodexVerifiedInstall
contact-widgetnexu-io/open-designSelf-contained floating chat widget with welcome screen, social links, meeting button, and message input. Single HTML file, zero dependencies.Stars: 108kUpdated Claude CodeCodexVerifiedInstall
meeting-notesnexu-io/open-designMeeting notes page — title bar with attendees, agenda checklist, decisions block, action items table with owners + dates, and a "next meeting" footer. Use when the brief mentions "meeting notes", "minutes", "1:1 notes", "all-hands recap", or "会议纪要".Stars: 108kUpdated Claude CodeCodexVerifiedInstall