What is your goal?
I am trying to connect to the Luma Dream Machine API to generate a video using the HTTP (Make a request) module.
What is the problem & what have you tried?
I am consistently getting a 500 Internal Server Error instead of a 400/401, even though my setup seems correct.
What I have tried to ensure it’s not a Make setup issue:
Verified my API Key is active and my account has a $5.00 balance.
Ensured the Headers field in the HTTP module is completely empty to prevent any “Authorization” header duplicate conflicts.
Stripped the JSON payload down to the absolute bare minimum to rule out formatting errors: {“prompt”: “A beautiful sunset over the ocean”, “model”: “ray-3.2”}. (I also tested with “ray-2”).
Could this be an issue with Luma’s API server, or is there a hidden HTTP module setting I might have missed?
Error messages or input/output bundles
Endpoint URL: POST https://api.lumalabs.ai/dream-machine/v1/generations
Error Message:
JSON
{“code”:-1,“message”:“Server Error”}
Code: ConnectionError
Screenshots (scenario flow, module settings, errors)
Emptying the Headers field to avoid a duplicate Authorization header may be the cause rather than the cure. If nothing else is injecting that header, you are now sending an unauthenticated request, and plenty of APIs answer that with a 500 instead of a clean 401. Put Authorization back in with Bearer and your key, plus content-type and accept as application/json, and see whether the status code changes.
Two more things before you suspect the HTTP module. Check the model string against the current Luma documentation, because an unknown model is another common source of a server side error rather than a clean validation error. And send the identical request once from outside Make, same URL, same headers, same body. If it fails there too, you have your answer and the module is not involved.
One practical note for debugging this kind of thing in Make: turn off Parse response on the HTTP module while testing. You then see the raw body exactly as the API returned it, including error detail that gets swallowed when Make tries to interpret the response.
And why are you not using the built-in Luma AI modules?
Hey KH_Chang! @core.Iq_GmbH is spot on about the headers. Just a quick heads-up: once you fix the auth, you’ll likely hit Make’s timeout limits because video generation APIs (like Luma) take time to process. You usually need a webhook or polling setup to catch the finished video without freezing the scenario.
Good point on the timeout, that is the next wall after the auth. The pattern that works: one scenario submits the generation and stores the returned id, then either a second scenario on a schedule polls the status endpoint until the state is completed and writes the file away, or you register a callback if Luma offers one and let a webhook catch it. Do not keep the original scenario waiting, it will die before the video is ready.
And on the built-in modules, those are the right default and I would start there. The HTTP module is for the cases where a parameter you need is not exposed, or where you want to see the raw response while debugging exactly this kind of error.
