Rendering and troubleshooting

When a job behaves unexpectedly, this page is the diagnostic ladder. Render queue states, what each means, what to do when one stalls, and the common error messages.

The job statuses

Every job moves through a fixed set of statuses. The dashboard's status pill shows the current one. In order:

1. draft. Brief submitted, queued for outline. Usually under a second.
2. outlining. Claude is generating the scene-by-scene outline. 20 to 40 seconds.
3. outlined. Outline ready, queued for script. The outline page is now readable; you can step in and regenerate sections if needed.
4. scripting. Each scene's script is being generated. Roughly a minute per minute of video.
5. scripted. All scenes have script text, queued for critic.
6. critiquing. Script critic running, flagging blockers / majors / minors. 30 to 90 seconds. May loop into a rewrite pass if blockers are found.
7. paused_for_review. A quality gate triggered. The dashboard surfaces a chooser panel. See "Paused for review" below.
8. ready_to_render. Your preview is composed (overlays, music, the lot) and waiting. Open it in the Studio, scrub it, tweak anything, then hit Render to produce the final video. Nothing renders until you say so, so you're never billed render time for a video you haven't seen.
9. rendering. Shotstack is stitching the final video from the timeline you approved. 3 to 8 minutes for a typical 12-minute video.
10. rendered. The MP4 is ready. The publish surface is now waiting.
11. published. Live on YouTube (or downloaded as MP4 if you went that route).
12. failed. Something went wrong end-to-end. The job-status page surfaces the error + an action button.
13. cancelled. You cancelled the job from the dashboard or editor.

The pipeline runs in parallel where it can. Footage and voice run concurrently with one another, both off the same script. Once they finish, the whole video is composed into a faithful preview and parked at ready_to_render. The final render only starts when you open that preview and click Render.

Live status on the dashboard

Each in-progress job shows a live-pulsing dot + the current status label. The dashboard live-updates without a refresh (live status is via Inngest); the rotating "behind the scenes" flavour copy changes with the stage so you know what the pipeline is actually doing.

If your tab has been open for a while and the status looks stale, hard-refresh once. The pulsing dot resumes within a second.

Paused for review

Four gates can pause a job for review. Each surfaces a chooser panel on the dashboard with named options.

Brief validator pause

The validator refused generation because something in the brief looks likely to produce a bad video. Common categories: hyper-specific visuals stock can't deliver, live-news content the pipeline doesn't do, command-prefix language that reads like a prompt-injection attempt.

The chooser is: refine the brief and re-run, or submit anyway. "Submit anyway" overrides the validator and the pipeline proceeds; use sparingly.

Footage availability pause

Fewer than 70 percent of scenes have a vision-accepted candidate. The orchestrator's chain ran but the pool came back thin.

The chooser is three cards: refine the brief and re-run, manually swap footage on the weak scenes via the editor, or accept the downgrade and proceed with the keyword-ranked fallbacks. Refining credits are free for the re-run (you don't pay twice for a brief tweak).

VO anomaly pause

The vision evaluator's audio check detected an artefact on one or more scene voiceovers (cutting, slurring, dropped words).

The chooser is: regenerate the flagged scene's voice, regenerate all flagged scenes, or accept and continue.

Content policy pause

The script or footage flagged against the Acceptable Use Policy. The chooser depends on the category. Usually it's refine the brief or edit the script by hand.

When a job is failing

If the status pill is failed, the job-status page shows the error category and an action button. Common categories and what to do:

"Claude rate limit"

Anthropic API rate-limited the request. The pipeline auto-retries with backoff up to three times; if all three fail, the job lands here. Action: Retry job (the button), which queues fresh after the retry window. Rate limits resolve within a few minutes.

"ElevenLabs synthesis failed"

Usually a transient ElevenLabs issue. Action: Retry job. If it fails twice in a row, drop the job ID to hello@cuethescene.com and we'll dig in operator-side.

"Shotstack render failed"

The render service returned an error. Most common cause is a transient Shotstack queue issue. Action: Retry render (this skips outline/script/voice/footage, just resubmits the timeline to Shotstack). Costs no fresh credits; the render-stage charge already applied.

"Footage source unavailable"

One of the niche's footage providers returned an error or zero results. The orchestrator usually falls through to the next provider in the chain; if every provider on the chain fails, the job lands here. Action: Retry job, or open the editor and use Search more with a different tier toggle.

"Brief validation failed"

The brief validator caught something post-submission (rare; usually catches at submit time). Action: Refine brief to fix the flagged category.

"Insufficient credits"

Your credit balance dropped below the job's reserved amount during processing. The job pauses, not fails. Action: Top up with a pack purchase or enable auto-PAYG.

"Unknown error"

The catch-all. Action: Retry job first; if it fails twice, drop the job ID to hello@cuethescene.com.

When a job is stuck

"Stuck" usually means the status pill hasn't changed for more than 15 minutes when you'd expect it to. The diagnostic sequence:

1. Hard-refresh the dashboard. Live status sometimes lags behind actual state by a few seconds; a refresh usually shows the real status.
2. Check the rotating flavour copy. If it's still changing, the pipeline IS running, you're just waiting on a slow API.
3. Check your concurrency. If you have your plan's concurrency limit of jobs in flight (1 on Starter, 2 on Creator, 3 on Pro, 5 on Scale), additional jobs queue rather than run. The status pill shows "queued" rather than the active stage.
4. Check the audit log on the job-status page. Every stage transition is logged. If the log shows no transitions in the last 15 minutes for an in-progress job, the pipeline genuinely is stuck.
5. Drop the job ID to hello@cuethescene.com. Real stalls are rare but they happen. Operator-side recovery is usually a manual Inngest retry, which we do within a few hours.

Cancelling a job

Cancel from the dashboard via the project row's "..." menu, or from the editor via the "Cancel job" button.

Cancellation refunds:

• Cancelled before outline lands. Full credit refund.
• Cancelled after outline but before script. Refund minus the outline-stage cost (10 credits for dry-run, or 100 + 50/min × outline-fraction of full).
• Cancelled after script but before render. Refund minus script + voice costs already incurred.
• Cancelled after render starts. Render is not refundable once Shotstack has started; the rest of the budget is.

The refund posts to your credit balance immediately. No support email needed.

Re-rendering an existing video

Re-render (from the editor's "Re-render" button) only re-runs the Shotstack timeline step. Outline, script, voice, footage remain as they were. Useful when:

• You edited captions and want them in the final MP4.
• You changed the music or audio mix.
• You changed a scene's footage candidate and want the new clip in the render.

Re-rendering costs the render-stage portion of the credit budget. Typically 80 to 120 credits for a 12-minute video, surfaced as a confirm dialog before you commit. See Plans and pricing for the full credit economy.

When YouTube publish fails

Three usual causes:

1. OAuth token expired. Visible as "YouTube connection lost" on the publish surface. Reconnect from Settings → Integrations.
2. YouTube category mismatch. Some categories have additional verification requirements. The publish surface surfaces the YouTube error verbatim; usually the fix is to pick a different category.
3. Video file size exceeds your channel's per-upload limit. YouTube's default per-upload limit is 256 GB; you're well under that on any CueTheScene render. If you hit a smaller per-channel limit, you'll see it in the error message.

If none of those match, drop the job ID to hello@cuethescene.com.

What's next

Editing your video covers the editor side of everything that happens BEFORE rendering.

Editor overview covers the editor's panel layout and the keyboard shortcuts.

Auto vs Manual mode covers when to let the pipeline run end-to-end versus stepping in at one of the paused-for-review gates.

Cheers,
Carl