CapCut AI Video Stuck at 99%: What to Check

CapCut AI Video stuck at 99%? Identify the active step preserve the exact message and choose a safe check before you retry or contact support.

Stuck at 99% cover with a paper checklist and a magnifying glass
CapCut
CapCut
Sep 10, 2026

When an image to video ai workflow appears stuck at 99%, start by identifying the operation behind the indicator. Record the exact message, surface, input category, and time, then choose the branch that matches what you see before changing the request. CapCut's public Help guidance groups troubleshooting around network conditions, service load, input complexity, and temporary disruption. Use those categories to choose the next check while the visible message and operation type guide the branch.

Phenomenon: confirm what is actually stuck

The first check is the operation, not the number. A near-complete indicator can be confused with another part of the workflow:

Visible state
What it may represent
What to record
An AI action shows a processing or thinking state
Cloud AI processing
The action name and exact status text
A timeline preview or render is updating
Normal editing or rendering
The panel or control that started it
A file is moving into the project
Upload or input transfer
The input type and any upload message
An export dialog is preparing a file
Final export
The export message and selected destination

CapCut's Help guidance on a stalled "Thinking" state describes this status as part of AI-powered operations and separates it from regular editing and export. Treat 99% as a location in the workflow, then use the operation name and exact message to select the next check.

Known action: save the smallest useful evidence

Before refreshing, cancelling, replacing an input, or submitting another request, capture a sanitized note. Do not include a private project name, email address, or other personal details.

Surface and URL: Action that started the indicator: Exact visible message: Input category: text / image / script / multi-scene brief Page still responds: yes / no Time recorded:

If a screenshot is needed, crop or redact private project information. Keep the original wording because "processing," "couldn't generate," an availability notice, and a policy notice each lead to a different check.

A physical checklist for diagnosing a stalled AI video task

Possible boundaries: follow the symptom branch that matches the message

Use the branch that matches the visible state. Do not infer a cause from the percentage alone.

The page still responds and the message has not changed

This remains a processing symptom. Check the connection status and look for a recovery control. If the page offers no recovery control while the state stays unchanged, keep the note and move to the support bundle instead of altering several conditions at once.

A requirement or availability notice appears

This is an access or feature-availability branch rather than proof of a processing failure. Save the exact notice and the surface where it appeared. CapCut's Help guidance on AI availability notes that access can vary by region, feature requirements, and surface. Do not keep resubmitting the same request when the interface is asking for a requirement, purchase, or supported-region condition.

The message refers to the input or request

Treat this as an input-validation, complexity, or policy branch. Preserve the input category and the wording that triggered the notice. For an image to video ai request, change one input variable only when the interface offers a clearly permitted controlled check and use material whose rights and sensitivity are already clear. Do not try to evade a policy control by disguising the request, and do not re-upload sensitive material only to reproduce the symptom.

The state changes to a generic failure

The problem is no longer a 99% symptom. Keep the original note, add the new error text, and route the case to the matching AI video help guides rather than treating the two messages as the same failure. If the message is a generation failure rather than a processing stall, use the relevant ai video generator troubleshooting path instead of repeating the same request.

The page stops responding

Preserve the last visible state and the time it was recorded. Avoid claiming that the job finished or that a usable video exists until the interface shows a result that can be reviewed. If the page later recovers, compare the new message with the saved note instead of relying on memory.

Known actions: run only the minimum safe checks

When the page provides a recovery control and the input is safe to use, keep any check controlled:

    1
  1. Read the current message and any requirement notice in full.
  2. 2
  3. Check the connection while leaving the input and other conditions unchanged.
  4. 3
  5. If input complexity is explicitly implicated, simplify one element and record what changed.
  6. 4
  7. Recheck the intended image to video ai workflow on the CapCut AI video tool before changing the request again.
  8. 5
  9. Stop if a policy, purchase, feature, or confirmation notice appears.

Treat one controlled check as a diagnostic observation. If the message changes, record both states and use the difference to choose the next branch; the percentage alone is less useful than that context.

Stop conditions

Stop the troubleshooting loop when any of these conditions applies:

  • The interface offers no recovery control while the page remains in the same state.
  • The same processing symptom returns after a controlled check.
  • A region, feature, purchase, or other requirement notice appears.
  • A policy or rights warning appears.
  • The page becomes unresponsive or the visible state cannot be recovered safely.
  • Reproduction would require uploading private or sensitive material again.

Escalation material

For support, send the exact message, CapCut surface, recorded time, input category, page-response state, and a sanitized screenshot if available. Omit personal account details and private project content unless support specifically requires them through its approved channel.

How to read the 99% signal

Read 99% together with the operation name and visible message. A server, network, input, policy, or completion branch needs its own visible evidence, so save the message, input category, response state, and time before choosing the next action. Keep the request unchanged for the first check, then change one variable only when the interface gives you a clear recovery path.

The next action is to save the triage note before changing anything. If the state is a different failure, use the AI video help guides linked above to choose the matching branch.

Hot and trending