Supervised mass generation, over HTTP

Render a small photo batch inline (synchronous, unsupervised)

Renders frames and returns them in the response. There is no job id and no QA: use this only for batches small enough to finish inside one request. continuityMode="chain" (the default) renders one frame at a time, conditioning each frame on the previous frame's output, so its ceiling is much lower than the parallel modes. Requests above the ceiling are rejected with 413 BEFORE any credit is charged and carry the durable path in durable_alternative. For genuine mass generation, or for anything that must meet a quality standard, use the supervised path: storyboard-images/plan then storyboard-images/generate. Feature-flagged (STORYBOARD_PHOTO_BATCH_ENABLED).

Compile the authoritative work plan for a supervised image batch

Returns the single authority for what will be generated, what it will cost, and what is missing: every required frame, the reference plates it depends on, the resolved provider, the credit ceiling, and any blocker that would stop execution. Spends no credits. Call this before generate to see the full cost and the blocker list.

Start a durable, supervised image batch

The supervised mass-generation entry point, and the one to build on. Persists a job before dispatching, so the batch survives client disconnects and process restarts. Execution is lease-exclusive: a duplicate delivery of the same task cannot make one job run twice. After rendering, every frame is scored against its requirements and failures enter a bounded repair loop with a per-requirement attempt limit, a per-shot credit ceiling and an escalation ladder that ends in human review rather than unbounded retries. Requires expectedExecutionManifestArtifactHash: call plan first and copy executionManifestArtifact.artifactHash from its response. Without it this returns 400 storyboard_image_execution_manifest_required — the hash pins execution to the exact plan that was priced, so a stale plan cannot be executed silently. Returns 202 with a job to poll (job.id, then GET film-intelligence/jobs/{jobId}) — unless every selected requirement was already satisfied by a prior call, in which case it returns 200 with accepted=false and no job: check accepted before assuming a job started.

Poll a durable generation job

Authoritative job state. status is queued, running, completed, partially_completed or failed. completed means every required frame was rendered AND passed review; partially_completed reports which requirements still failed and whether the batch is resumable. Safe to poll from any client, including one that did not start the job.

Read a photo sequence plan and its current lifecycle

The plan, its frames and where it currently stands. Spends nothing. Use it to price a repair before authorizing one, and to see the lifecycle without mutating anything.

Record a human accept/repair decision per frame

This is the human authority step, NOT the automatic scorer. Automatic scoring runs inside the generation job, and its verdicts and evidence are read from film-intelligence/jobs/{jobId} — not here. This endpoint records what a person decided about each frame, so it requires a decisions array (frameIndex, accept|repair, a note, optional prompt additions) and an existing QA report for the plan. Calling it without decisions returns 400, and with no prior QA report returns 409.

Repair only the frames that failed review

Regenerates the failing requirements and nothing else, so accepted frames are frozen rather than re-rolled. Bounded by attempts per requirement, requests per shot and a credit ceiling; each escalation tightens constraints before falling back to a frozen prompt, and exhausting the ladder returns human_review instead of spending more.

What this deployment can actually execute right now

Runtime capability truth, not documentation. Each capability reports configured plus the honest limitations that apply, so a client (or an agent) can check whether a workload is executable before spending anything. Unconfigured providers are reported as unavailable rather than failing at dispatch.

Convert text to a neural voice

Text-to-speech (legacy alias)

Generate a personalized commerce video (abandoned cart / post-purchase)

Generate a short vertical LTX video for a shopper from a commercial event. trigger=abandoned_cart produces an objection-handling video; trigger=post_purchase produces a cross-sell video. Send dryRun=true to preview the resolved brief + payload without spending credits. LTX only.

Get current user credit summary (stripe + bonus + total)

Deliver a finished video to agnostic destinations (webhook / email / Drive)

Delivers an already-rendered video; spends NO generation credits. Webhook is fully wired (SSRF-guarded + HMAC-signed via X-MYEQ-Signature); email and google_drive return an honest "unsupported" until their adapters land. Feature-flagged (UNIVERSAL_DELIVERY_ENABLED).

Generate a static AI image

Compose an original AI song

Generate music via async orchestration pipeline

Check generated music status by job id

List user projects

Create a project

Resolve final downloadable URL for a project or orchestrator job

Get async job status and progress

Proxy an allowlisted image preview URL

No open proxy: only HTTPS URLs from approved hosts are fetched; Wasabi objects are signed with a one-hour TTL.

Get a signed URL for direct S3 upload

Issues a short-lived upload URL under the authenticated user prefix. Category limits: PHOTOS 20 MB, VIDEOS 500 MB, MUSIC/SFX/VOICES 100 MB.

List assets available in Creative Hub library (videos, photos, music, voices)

Get user creative-agent personalization settings

Update user creative-agent personalization settings

Create a spend-gated video generation job

Backward-compatible alias for /api/v1/video/create. This endpoint may start paid generation only after successful authentication, request validation, deterministic preflight, provider capability validation, and credit/BYOK authorization. LTX lip-sync preflight failures return 422 with no_credits_spent=true and no provider dispatch.

Create and start a long-video job

Create an orchestrated multimodal video job

Run AI quality director review for a video project

Check generated video status by job id

List accessible workspaces

Create a team workspace

Delete/detach a non-protected workspace

Requires owner membership. Personal/default and last remaining workspaces are protected. Assets are not permanently deleted by this endpoint.