Cinely

AI Video Stuck at 99% or Failed? Get Credits Back

Cinely Team··9 min
A film reel frozen mid-spin with its film strip snapped, representing a stuck AI video

The progress bar reached 99% twenty minutes ago and hasn't moved. Or the spinner finally stopped on "Generation failed," and your balance is lower than it was before you pressed the button. Now you're deciding whether to wait, retry or complain, and wondering if those credits are gone.

Here is the short answer. An AI video stuck at 99% is usually waiting on a step the progress bar can't see, or it has already died on the server. Give it a fixed amount of time, look for the finished file in your library and on the website before you retry, screenshot your balance, then ask support for the credits from that one job. A store refund is the last step, not the first.

Failed jobs are one of the quieter ways credits disappear. For the bigger picture, see how AI video generator credits get used up.

What are people reporting about stuck and failed AI videos?

In September 2026 we read more than 12,000 one- and two-star reviews of 63 apps on the App Store, Google Play, Trustpilot and Capterra. Among AI video generators, stuck and failed jobs came up constantly, in three shapes.

The job never starts. Pika's App Store reviewers describe queues of one to five days, including on paid plans. A Kling AI video stuck for hours is a familiar story on Google Play too, where one widely upvoted review describes two- to three-hour waits for a five-second clip. On Sora's Android app, reviewers kept landing in an "under heavy load" loop before OpenAI closed the app.

The job never finishes. One Google Play reviewer wrote that on Dreamina "you will wait to 99% then it will fail 90% of the time." A Viggle reviewer said the app "estimated a 16-minute wait time. Instead, it took 25 hours to process," and the job still ended as failed. A HubX reviewer waited 12 hours for a video the app said would take two minutes.

The job fails and keeps the credits. About 14% of Google Flow's low-star Google Play reviews say failed generations still cost credits. A HeyGen reviewer on the App Store described a render that sat frozen: "Video processing hangs - stuck at 0% … but still eats credits." One Hailuo reviewer on Google Play says credits came off for each failed attempt until they couldn't try again.

Experiences vary, and these apps change often, but the causes behind all three are few.

Why is my AI video stuck at 99%?

Because the number on the bar often isn't a measurement. A few different things can be happening behind it.

The bar is a guess. In a lot of apps, the bar moves on a timer based on how long jobs usually take, then waits near the end for the server to report back. If the job runs long, the bar parks at 99%. Kling reviewers even describe progress bars that jump backwards.

You're in a queue. Video models run on expensive shared hardware, and there's only so much of it. When demand spikes, jobs wait their turn, and free plans usually wait behind paid ones. Pika reviewers describe a free tier whose queue never seems to move unless you pay to skip it.

The provider is overloaded. Many apps don't run their own video model. They pass your request to a model provider, and when that provider is busy, the app can only wait or give up.

The last steps broke. After the frames are drawn, a clip still has to be encoded, often checked by a safety filter, and uploaded before you can play it. A video made of several clips also has to be joined. A failure at that stage leaves a job that looks nearly done and never arrives.

Your phone lost track. Lock the screen or switch apps, and the phone may pause the app's connection, so it misses the "done" signal even when the server finished. Viggle reviewers say its app has to stay open while it generates, and some saw network errors on strong Wi-Fi.

The server gave up. Jobs have time limits. One Hailuo customer on Trustpilot described waiting hours for a video that then timed out with nothing to show for it.

How long should an AI video take?

There's no single right number, because the model, the length and the service's workload all change it. On a healthy service a single short clip usually takes minutes, and a video built from many clips takes longer. When reviewers report 12 hours or 25 hours, that usually points to a long queue or a job that has died.

A workable rule: wait twice the app's own estimate. If it gives none, allow about 30 minutes for a single clip and an hour for a multi-scene video before you treat it as stuck. Meanwhile, leave the job alone, and keep the app open if it's one that needs to stay in front. For everything else that shapes a project's timeline, see how long it takes to make an AI movie.

Did you lose your credits when it failed?

Not necessarily. Apps handle failed jobs in one of three ways, and the Generate button won't tell you which:

  1. Charge when you start, refund automatically. Credits leave on submit and come back on their own if the job fails.
  2. Charge when you start, refund on request. Credits leave and only return if support restores them.
  3. Charge only on success. Nothing leaves your balance until a video is delivered.

If your AI video failed but took credits, check the app's credit or usage history, not just the error message. Google Flow reviewers say failed generations were charged even when the app said they weren't. And don't delete the stuck or failed job. It's your record of what happened.

How do you get credits back for a failed generation?

Work through these steps in order. Skipping straight to a retry is how people end up paying twice.

  1. Wait out your window. Use the rule above. Late finishes are common, and a job you gave up on may still complete.
  2. Don't regenerate yet. A retry is a new charge, and if the first job finishes after all, you've paid for the same video twice. Retry once the app shows a refund or support confirms the job is dead.
  3. Look where finished videos land. Check the library, history or drafts tab, your notifications, and the website if there is one. Apps and servers don't always agree. One PixVerse reviewer on Google Play put it bluntly: "If you HAVE to use Pixverse, use the web browser instead of the app."
  4. Reopen once. Close and reopen the app, or reload the page, then check your balance and history again.
  5. Screenshot everything. Use the table in the next section.
  6. Ask support for that job's credits. Name the job, the time and the amount charged. A small, specific request is easier to approve than a general complaint.
  7. Escalate only if support stalls. An AI video refund from Apple, Google or your card issuer returns money for a purchase, not credits for one job. It makes sense when failures kept you from using what you paid for and support won't respond. Our guide to refunds on the App Store, Google Play and the web covers each route.

If the video wasn't stuck but blocked by a safety filter, resending it won't help. Here's why harmless prompts get flagged and how to rewrite them.

What should you record before contacting support?

Support can only restore credits it can find. Capture these before you retry anything:

What to captureWhy it helps
Your balance before and after the jobProves the charge happened
The job entry: time, prompt, length, and ID if shownLets support find the exact render
The error text, or the frozen progress screenShows a failure, not a filter block or a slow queue
How long you waitedShows you didn't give up too early
Web or app, your device, and the app versionHelps them match a known bug
Any "not charged" or "refunded" messageHolds the app to what it told you
Your ticket number and the date you wroteYou'll need it if you escalate

Then write three or four lines: what you made, when, what happened, what was charged, and that you want those credits restored. Asking about an AI video generator refund of your whole plan in the same message only slows the answer down.

Failures belong in your budget, too. Our breakdown of what an AI movie really costs per finished minute shows how retries and fixes change the price.

What happens on Cinely when a scene fails?

Cinely is an AI movie maker for the web, iOS and Android. You describe an idea or paste a script, and it renders each scene as an 8-second clip, then joins the scenes into one film. That keeps a failure contained: one bad scene doesn't have to sink the movie.

While it renders. The website says "This may take a few minutes" and gives no time estimate, which is a real limitation: you can't apply the wait-twice rule there. You don't have to watch, though. "Go to Library" lets you leave the page, and when the movie completes, Cinely creates a "Your movie is ready!" notification that also arrives as a push in the Cinely mobile app. The app's progress screen says "Typically 3–5 minutes" and that you can close the screen. Treat that as a typical case, not a promise.

If the whole movie fails. Credits charged for it are refunded automatically, and the failure screen says "Your credits have been refunded." That line only appears when the server confirms the refund actually happened, so a failure that never charged you doesn't promise one. A film stopped by the safety filter takes the same path.

If some scenes fail. The movie still finishes with the scenes that worked, and you see "Some scenes could not be generated and were refunded." In the editor, each failed scene shows as "Missing scene," and you can regenerate it back into the episode.

Other jobs. Failed editor fixes, thumbnails and Cinematic remakes are refunded too, and a background job retries every minute any refund that didn't go through.

What changes with Premium. Premium's default Pro mode costs the same 30 credits per scene and adds fallback: if the first attempt fails, a second model stage and a backup provider can take over. Free Standard tries its model up to three times, with no fallback.

What isn't refunded. Refunds cover failures, not taste. A scene that rendered but that you don't like keeps its charge, and fixing it in the editor costs 60 credits. More answers are in the Cinely FAQ.

Next time a bar sits at 99%, give it a set window, find the job before you retry, and screenshot your balance. If you'd rather not chase credits at all, start your next film on Cinely, where failed scenes are refunded without a support ticket.

Should I delete a stuck AI video and start over?
Not yet. The stuck job is your best evidence, and deleting it rarely gets credits back: Luma reviewers say deleted pending generations still cost them. Leave it alone, wait out a reasonable window, then check your library and the website. If it is still stuck, screenshot the job and your balance, and ask support to cancel it and restore the credits before you start a new one.
Why is the AI video app slower than the website?
Some services run different limits or queues in their mobile apps. One Dreamina reviewer on Google Play said generations took seconds on a PC but 5 to 10 minutes in the app. Phones also pause apps running in the background, so a job that finished on the server can look stuck on your screen. If the app stalls, sign in on the website and look for the video there before you retry.
Does paying for a plan make AI videos render faster?
Not reliably. Pika reviewers describe queues lasting days on paid plans, one Hailuo reviewer on Trustpilot said its most expensive plan was slower than cheaper tiers, and Higgsfield customers with "unlimited" passes describe jobs stuck on processing. On Cinely, what Premium's default Pro mode adds is resilience: a second model stage and a backup provider for when the first attempt fails, at the same price per scene.
Can Apple or Google refund credits lost on one failed video?
They refund purchases, not single generations. If failures kept you from using a plan or credit pack you bought in the App Store or Google Play, you can ask for a refund there, but approval is not guaranteed, and a refunded purchase usually takes back what it granted. For credits lost on one job, the app's own support is faster. On Cinely, credits for failed renders come back automatically.

Written with AI assistance and edited by the Cinely Team.