Luma Dream Machine Not Working? — Troubleshooting
Troubleshooting

Luma Dream Machine not working? Here’s what’s actually going wrong.

Stuck queue, failed render, or a video that came back looking nothing like your prompt — nine specific causes and the exact fix for each, not “try again later.”

Queue

Generation stuck on “processing” for more than 10 minutes

Most free-tier queues clear in 2–4 minutes. Past 10 minutes, one of three things is happening:

  • Server load spike — most common on US evenings and weekends. Refresh once; don’t resubmit the same prompt, or you’ll queue a duplicate job behind the first one.
  • A silent failed job that never cleared — log back in and check your generation history. If it isn’t listed at all, it never started. Resubmit.
  • Browser tab throttling — a backgrounded tab can pause its status check. Bring it to the foreground and give it 30 seconds before assuming it’s dead.
Prompting

The video generated but looks nothing like the prompt

Almost never a broken-tool problem — a prompt structure problem. The model weights subject, action, and camera roughly in the order you write them. “A dog running on a beach at sunset, cinematic” hands it three unranked ideas to reconcile at once.

Restructure as subject → specific action → camera move → light/mood, one clause each:
“A golden retriever running along a beach. Camera tracks alongside at ground level. Warm sunset light, long shadows.”

That’s the single biggest quality jump available, and it costs nothing to test.

Output quality

Hands, faces, or text are warping or melting

A known limitation of the current model generation, not a bug tied to your account. Shows up most on:

  • Close-up shots of hands doing something specific (holding an object, playing an instrument)
  • Any legible text in the scene (signs, labels, screens)
  • Faces at an extreme angle or in fast motion

Workaround: keep hands and faces at medium-to-wide shot distance, and avoid prompting for visible text — it will almost always come out garbled.

Plan limits

Video generated with a watermark you didn’t expect

Free-tier and unpaid-plan generations carry a watermark by default — that’s an expected limit, not a bug. If you’re on a paid plan and still seeing one, check that the plan is actually active on the account you’re logged into, not a second account or a session that quietly logged you out to guest.

Failed jobs

“Generation failed” with no video and no credit refund

Save the exact prompt text and rough timestamp before doing anything else — you’ll need it if you contact support. Then check:

  • Did the prompt include anything that could trip a content filter (real public figures, brand names, violent or explicit description)? Filtered prompts often fail silently rather than warning you upfront.
  • Was the source image (for image-to-video) a reasonable file size and a standard format (JPG/PNG)? Oversized or unusual formats are a common silent-failure cause.
Plan limits

Free plan generation limit reached mid-project

Free tiers reset monthly, not daily — hitting the ceiling mid-month means waiting for reset or upgrading, with no way around it on the free tier itself. If you’re mid-project and it can’t wait, a short-term paid upgrade is usually cheaper than losing momentum on a deadline.

Downloads

Video downloads but won’t play or looks corrupted

Almost always a partial download, not a corrupted generation. Re-download rather than troubleshooting the file itself — if the tab closed or lost connection mid-download, you’ll get a file that opens but stutters or cuts off.

Prompting

Prompt keeps getting auto-modified before it generates

Some platforms run your prompt through an “enhance” step automatically, rewriting it before generation. If your output consistently doesn’t match what you typed, look for a toggle to disable prompt auto-enhancement — usually on by default, and it works against you once you’re writing detailed prompts yourself.

Support

Still stuck after trying the above

Note the exact prompt, the time (with timezone), and whether you were on free or paid tier — send all three upfront. That cuts the back-and-forth down to one message instead of four.

Found a fix that isn’t listed here?

Tell us what worked and we’ll add it — this page only gets more useful with more real cases.

Contact us