Troubleshooting & FAQ

Sometimes a step in your workflow doesn't work the way you expect. This page explains what the error messages actually mean and what to do about each one.

How errors show up

When a step in your workflow fails, a popup appears on screen with a warning icon. It tells you which step failed and gives you a plain-English reason, so you don't have to guess what went wrong.

If there's more detail available, you'll see a Technical details section you can click to expand. This shows the raw message from the provider (with anything sensitive removed), in case you want to share it when asking for help.

Common error messages and what to do

"Add your Atlas Cloud API key in Settings → API Keys." You haven't added your Atlas Cloud key yet. Go to Settings > API Keys and add it.

"Your <Provider> API key was rejected. Check it in Settings > API Keys." The key you saved isn't valid, either it was typo'd, revoked, or expired. Go to Settings > API Keys, remove the old key, and add a fresh one.

"Your Atlas Cloud balance is too low. Top up at atlascloud.ai, then run again." Your Atlas Cloud account has no credit, either because it ran out or because you haven't topped it up yet. Atlas Cloud runs nothing on an empty account. Add credit on atlascloud.ai, then run the workflow again. On the canvas and in Creative Studio the message is a link. Click it to open Atlas Cloud's billing page in a new tab.

"<Node name> produced no output. Try a different prompt or model." The model returned an empty reply with no specific reason given (a separate message appears when the safety filter is the identified cause, see below). Try rewording your prompt or switching to a different model.

"LLM: no prompt yet. Type a prompt in the node settings, or connect a text-producing node to the Prompt input." Your LLM node has nothing to work with. Either type a prompt directly in the node, or connect a node that provides text to its Prompt input.

"LLM: the Prompt input is connected, but no text arrived." Something is wired into the Prompt input, but it handed over no text. Check the node that feeds it, or remove the connection and type a prompt instead. Image Generation, Video Generation and Voice Generation show the same kind of message for their own inputs.

"This request was blocked for possible copyright or likeness reasons. Remove named people, brands, or trademarked characters and try again." The provider flagged your prompt as referencing a real person, brand, or trademarked character. Rewrite the prompt without those references.

"Your content was flagged by the safety filter. Try a different prompt." The provider's safety filter blocked your request for another reason (for example, NSFW or policy-restricted content). Adjust your prompt and try again.

"<Provider> rate limit reached. Try again in a few minutes or reduce concurrent workflows." You're sending requests to that provider faster than it allows. Wait a few minutes, or run fewer workflows at the same time.

"Atlas Cloud is busy with this model right now. Try again in a few minutes or pick another model." Atlas Cloud accepted your request, but could not run that model at that moment because too many requests were going to it. Nothing is wrong with your prompt or your files. Wait a few minutes and run the node again, or switch the node to another model.

"Image is too large (max 2048x2048). Use a smaller image." (or similar, for video/audio) Your uploaded image is larger in dimensions than the model supports. Resize it to fit the stated size and try again. When the provider tells us the exact limit, we show it in the message.

Unsupported file format The file type you uploaded isn't supported by that model. The message will list which formats are supported when that information is available.

"<Node name>: the prompt is too long for this model. Shorten it and run again." Your prompt has too much text for the model's limit. Trim it down and run again.

"...the workflow will retry automatically." or "...Try running the workflow again." The provider had a temporary problem on their end. Some failures like this are retried automatically; others need you to hit run again.

"Generation took too long. Try a simpler prompt or different model." The request took longer than the provider allows. Simplify your prompt, or switch to a faster model.

"<Provider> could not find the requested model or resource. Choose a different model." (exact wording varies by provider) The specific model or resource this step was pointed at is no longer available. Pick a different model in the node's settings.

"The run was cancelled." You (or something you triggered) stopped the run before it finished. This isn't an error. It's just confirming the run was cancelled.

"This step stopped unexpectedly. Run the workflow again." The step ended without reporting a result, so the run was marked failed instead of staying on "running". Run the workflow again. If it keeps happening on the same step, contact support.

"This step was stopped." The step was stopped before it could report a result. Run the workflow again.

"No model was saved on this node, so it ran on …" The node had an empty model setting, so a default model was used. Open the node's settings and pick the model you want.

"This workflow could not be found. It may have been deleted." The workflow you tried to run no longer exists.

"<Node name> has no saved data. Save the workflow before running it." You ran a workflow with unsaved changes on this node. Save the workflow first, then run it.

"The workflow has a circular dependency. Remove the loop so data flows in one direction." Two or more nodes are connected in a loop. Rework the connections so data flows in one direction, from start to finish, with no loop back.

"Start your subscription to run workflows. Your 7-day free trial begins at checkout." / "Your free trial has been used. Subscribe to keep running workflows." Running workflows needs an active 60 Minute Creator plan. You can still build and edit everything without one. Pick a plan from Settings > Billing. Re-running a node on results you already generated works without a plan and doesn't use up a run. Your provider still charges for what it generates.

"Monthly run limit reached (N). Buy a run pack or upgrade your plan." / "Free trial run limit reached (N). Your plan's full allowance unlocks when the trial ends and your subscription starts." You have used all the runs your plan includes for this billing period. Buy a run pack or upgrade in Settings > Billing, or wait for the next period. During a free trial you get a smaller allowance, and the full one starts when your subscription does. See Limits & Formats for the exact run limits per plan.

These four messages read the same wherever they appear: when you press Run, on a scheduled run, from your AI assistant, or in the run's history.

"<Node name> generated a result but saving it failed. Try running the workflow again." The step actually worked, but we couldn't save the result. Run the workflow again.

Anything else If a message doesn't match a known cause, you'll still get a clear one-line summary rather than a raw error, plus a note to try running the workflow again.

Checking your API keys before you run anything

You don't have to wait for a run to fail to check a key. Open Settings > API Keys and look at the status pill next to each provider:

• Valid: the key works.
• Invalid: the key was rejected. Remove it and add a working one.
• Not configured: no key has been added yet for that provider.

When you save a new key, it's checked against the provider right away. If it doesn't work, you'll see: "This key is not valid for <Provider>. Check that you copied the full key."

FAQ