Why AI Video Generations Fail (and What Happens to Your Credits)

An AI video generation fails for one of four reasons: a safety check stopped the prompt, the attached image or the result; a setting or file was not valid for the model you picked; the model was busy or broke on its side; or the job ran out of time. Knowing which of the four happened tells you whether to retry immediately or change one input first.

This guide covers what each failure message means in the AIGE studio, what to change before you run the shot again, and how credit charges work when a clip does not arrive.

Why did my AI video generation fail?

When a generation fails in AIGE, the card that replaces the clip names the reason. Nearly every failure belongs to one of these four families:

The family matters more than the wording, because it tells you whether to press generate again or to change something first.

Safety blocks: find out which part was stopped

A safety block is the most common reason a video never arrives, and "blocked" on its own is not enough to act on. AIGE separates four cases, because each has a different fix.

The prompt was blocked. A word or phrase set off the check. Reword that part and run it again.

The attached image was blocked. The check stopped on your start picture or reference, not on your wording. Rewriting the prompt will not help. Swap the image.

The result was blocked. Your prompt and image were accepted, and what the model produced crossed the line. Run it again, or describe the scene in safer terms.

The model returned nothing. Some models stop a request without saying why. That almost always means a quiet safety stop, so treat it as a prompt block and reword the riskiest part.

One case is different. A few subjects are hard limits on the platform, and those requests are never sent to a model at all. Rewording does not get them through. Requests built around a real, identifiable person fall here: describe the look you want instead of naming someone.

Settings and files the model cannot take

These failures are the easiest to fix, because the card usually names the control to change, for example "This model doesn't support the selected aspect ratio."

• Clip length. Each model renders its own range. In AIGE, Veo 3.1 makes clips of 4, 6 or 8 seconds, Kling V3 runs from 3 to 15 seconds, and Seedance runs from 4 to 15 seconds (2.0) or up to 30 seconds (2.5). The length selector only shows what the chosen model supports.

• Start picture too small or too large. Use at least 512×512 pixels, and stay under 4K.

• No visible face. Models that animate a person need a clear, well-lit face in the picture.

• File format. JPG, PNG or WebP for images, MP4 for video, MP3 or WAV for audio.

• Prompt length. Every model caps how much text it reads. If the prompt is too long, cut it down to the core idea.

If you start clips from a picture often, the short guide to image-to-video explains how the start and end frames steer the shot.

What happens to my credits when a generation fails?

How credits are handled depends on where the request stopped:

• Rejected when you submit. If a request is refused before the job queues (for example, an invalid setting or a request that fails to submit), the studio restores the deducted credits to your balance immediately.

• Generated from a chat assistant over MCP. When you run AIGE's MCP tools from Claude, ChatGPT or Cursor, each queued video task records its own credit receipt and restores the charge if the task fails or times out.

• Downstream failure after queuing in the web studio. If a queued clip fails later on the model's side and your balance was not restored automatically, email support@aige.ws with the prompt or timestamp so we can restore the credits, or request a refund of unused credits within 14 days under the refund policy.

What a refund does not cover is a clip that arrived and that you do not like. A delivered video is a completed generation, even when the motion is wrong. That is the reason to check the price before a long take: the studio shows the credit cost before anything runs, and the cost per clip page lists it for every model.

Why is my video stuck on processing?

A clip that sits on "processing" has not failed yet. Most clips arrive in about 30 to 90 seconds; long takes and busy hours take longer. Behind the card, two scheduled checks keep a stalled job from hanging forever:

• Recovery, every 3 minutes. If the model finished your clip and the step that saves it broke, a recovery pass finds the finished video and delivers it.

• Cleanup, every 10 minutes. A task that has made no progress for more than 15 minutes is marked as failed so the card clears instead of spinning forever.

Retry, or change something first?

If the same model fails twice in a row on its own side, switch models for that shot. Kling, Veo and Seedance sit in the same picker and accept the same prompt and start picture, and the cost per second comparison shows what the swap does to the price.

Does this change between Kling, Veo and Seedance?

Every video model in AIGE runs through the same task queue and error classifier, so failure cards work the same way across all three. What changes is which limits you can hit: the clip lengths above differ per model, some models accept an end picture and some do not, and each model's safety check draws its lines in slightly different places. A prompt one model stops can pass on another, which is a fair reason to try a second model before rewriting a whole scene.

Quick answers

Do this next

Open the failed card and read which family it belongs to. If it was the model's side or a timeout, run it again as it is. If it was a block or a setting, change that one thing first. Then open the AI video generator, draft the shot, and check the quote before you run it.