When a Creation Fails
A creation that fails in Scholarly is never a dead end. Instead of a spinner that stops or a red toast that disappears, you get a panel that names the cause in plain language, keeps whatever work it managed to finish, and offers only the fixes that actually apply to that cause.
You will see that panel in three places, and it is the same panel each time:
- On the artifact's own page, when you open the thing that failed.
- In the create window, if you were still watching when it stopped.
- From its row on Home, alongside your other recent and in-progress work.
What the reason means
| The panel says | What actually happened | What usually fixes it |
|---|---|---|
| Not enough content | The source did not carry enough material to build what you asked for — a two-line file, a mostly blank scan, a page range with almost nothing on it. | Add a fuller source, or widen the page range. |
| Blocked by a safety filter | Something in the source or the prompt tripped the model's content rules. | Remove or replace the source that triggered it. |
| Too large | The source exceeded what this creation can read in one pass. | Narrow to a page range, or split the document. |
| Unreadable source | The file uploaded but its text could not be extracted — a scanned image with no text layer, or a damaged file. | Replace it with a text-based version. |
| Provider error | The AI model itself returned an error. | Try again, or try another model. |
| Timed out | The job ran past the time allowed for it. | Try again with a smaller or narrower source. |
| At capacity | Demand was unusually high and the job could not be placed. | Try again in a few minutes. |
| Rendering, narration, or script step failed | The text came through but one production stage — building the visuals, generating the voice, or writing the script — did not finish. | Try again; this stage is usually transient. |
When the generator has something more specific to tell you, it says that instead. A sentence written for your exact failure replaces the generic one above — read it first, because it is the more useful of the two.
Why the buttons differ each time
This is the part worth understanding, because it saves you from retrying something that can never succeed.
Causes that live in your source reproduce on every attempt. A file that is too large is too large again. A scan with no readable text is unreadable again. A source that is too thin stays too thin, and a source that trips the safety filter trips it again. Running the same job a second time would burn your time for a guaranteed identical result — so for these, no bare Try again is offered at all. The panel instead steers you toward changing the inputs: narrow the pages, swap the file, or drop the source that caused it.
Causes that are momentary do offer Try again. A provider error, a timeout, a capacity problem, a rendering or narration or script step that did not finish — none of these say anything about your material. The same sources very often succeed on the next run, so Try again is right there.
If you are ever unsure which kind you are looking at, the presence or absence of Try again is the answer.
The recovery actions
- Try again. Runs the same creation, with the same sources and settings, from the start. Offered only for transient causes.
- Try another model. Reruns the creation on a different AI model. Useful when one model refused, stalled, or returned an error on material another handles fine.
- Pick fewer pages. Opens the page selector so you can cut a long document down to the section that matters. Offered only when the run that failed used a paged source such as a PDF.
- Replace the source. Swaps out the file that caused the problem and keeps the rest of your setup.
- Adjust the sources. Reopens the source list so you can add, remove, or reorder what the creation is built from.
Retries respect your current plan and model access, exactly like a first attempt: a model your plan does not include is not an option on a retry either, and the source limits for your plan still apply.
What is kept
Scholarly tries hard not to make you rebuild a setup you already got right.
- Failed flashcard runs restore their sources and settings. Reopen the creation and your files, page ranges, and options are still filled in — adjust the one thing that was wrong and go.
- A deck keeps the cards it completed. When only some sections of a long source failed, the successful cards are kept rather than discarded, with a warning that coverage may be incomplete. Build a second deck from the remaining pages if you need full coverage.
- Video lectures keep generated narration across a retry. The voice work already produced is reused, so a retry after a rendering problem does not start from silence.
- Podcasts keep completed narration while waiting on a busy audio step. If the audio stage is backed up, the narration already recorded is held rather than thrown away.
Credits
- A creation that fails outright returns its AI Creation Credit automatically. You do not need to ask, and you do not need to count — the credit comes back on its own.
- Starting the creation again is a new creation. It draws on your current allowance: the week's credits on a paid plan, or your lifetime credits on Free, which do not reset.
- Repairs and whole-deck refinements to a flashcard deck you already have do not spend a credit. Fixing or refining an existing deck is not a new creation.
See Plans and Limits and AI Creation Credit Packs.
When it's a plan limit, not a failure
A plan limit is not the same thing as a failure, and it does not look like one. Nothing broke: you reached a ceiling.
The message names exactly what ran out — creation credits, uploads, chat messages, quiz questions, exam attempts, file size, or PDF pages — and shows the upgrade that lifts it, right where you hit it. Your existing work is untouched. Nothing is deleted, nothing is downgraded, and everything you already made stays open and usable.
Problems caught before you start
Some problems never get as far as a failure, because Scholarly checks for them at the moment you add the source and tells you immediately:
- An empty file — nothing to read, caught before you wait on an upload.
- A password-protected PDF — Scholarly cannot open it; remove the protection and upload it again.
- An Office lock file — the small placeholder file Word or PowerPoint leaves behind while a document is open, rather than the document itself. Upload the real file.
- A YouTube video that is deleted, private, region-blocked, or still live — each is named for what it is, so you are not left guessing. Wait for a live stream to end and be published before using it.
- A saved web page with no real content — a page that saved as navigation and boilerplate with no article behind it.
More of these, with the exact wording you will see, are in Fixing Common Problems.
Related
- Fixing Common Problems — the full catalogue of messages and what to do about each.
- Background Tasks and Notifications — why closing the tab is safe, and where to watch progress.
- Plans and Limits — the ceilings a limit message is referring to.
- Choosing an AI Model — what Try another model is choosing between.
- Supported File Formats — what Scholarly can read before you upload it.