You uploaded your song, wrote a creative direction, hit generate, and got an error instead of a video. Or worse: it ran, spun for a while, and still failed. AI video generation failed is one of the most common searches for a reason. Most failures trace back to one of five things: a file that does not meet spec, a song that is too short, a queue or credit issue, a prompt that confuses the pipeline, or a scene that needs a targeted fix rather than a full restart.
This is the checklist version of troubleshooting a failed generation. Work through it in order. Most artists find their answer in the first three sections.
Key Takeaways
- Most AI video generation failures come from five root causes: file format or size, song duration, credits or queue state, a malformed prompt, or a scene-level issue mistaken for a project-level failure.
- File format and size checks catch the majority of upload-stage failures. Echonos Engine accepts MP3, M4A, WAV, AAC, OGG, and FLAC up to 40 MB.
- Song length has a hard floor. Minimum 60 seconds; anything shorter fails at the audio analysis stage before generation even starts.
- Credits and queue status explain most "nothing happened" failures. Check your balance and job status before assuming the pipeline is broken.
- Not every bad result is a failed generation. A weak scene inside an otherwise working video calls for Studio scene regeneration, not a full re-run.
The Most Common Reasons a Generation Fails
Five categories cover almost every failed generation. Working through them in this order saves the most time, because the first two are the fastest to rule out.
1. The audio file does not meet the upload spec. Wrong format, file too large, or the file is corrupted or partially uploaded. This fails immediately, usually before you even reach the creative direction step.
2. The song does not meet the duration requirement. Too short, or in rare cases a file that reports a duration different from what actually plays (a metadata mismatch from certain export tools).
3. A credits or account state issue. Insufficient credits, a stuck job in the queue, or a network interruption mid-upload that leaves the job in a bad state.
4. The creative direction or prompt confuses the pipeline. Contradictory instructions, an empty or near-empty prompt, or a prompt written in a format the pipeline cannot parse into scenes.
5. The result is not actually a full failure. The video generated, but one or two scenes look wrong, and the instinct to call the whole thing "failed" leads people to restart from scratch when a scene-level fix would solve it in less time and fewer credits.
Each of the next sections works through one of these in more depth, since the checks and fixes differ by cause.
File Format and Size Checks (40MB, Supported Types)
This is the single most common reason a generation fails before it starts.
Echonos Engine accepts these audio formats: MP3, M4A, WAV, AAC, OGG, and FLAC. AIFF is not on this list. If your master was exported from a DAW as AIFF (common on Logic Pro and some Apple-centric workflows), re-export as WAV or a high-bitrate MP3 before uploading. This single swap resolves a large share of "upload failed" reports.
The maximum audio file size is 40 MB. A song mastered as a high-resolution WAV can exceed this comfortably, especially anything over four minutes at 24-bit. If your WAV is too large, either export a high-bitrate MP3 (320 kbps holds up fine for generation purposes) or use FLAC, which compresses losslessly and often lands well under the limit for a typical single.
For image uploads (character references, style references), the accepted formats are broader: PNG, JPG, JPEG, WebP, BMP, TIFF, TIF, SVG, HEIC, HEIF, and ICO. Character reference images have their own size limit: up to 10 MB per image.
A quick pre-upload checklist:
- Confirm the file extension is one of the six supported audio formats.
- Check the file size in Finder or Explorer before uploading, not after the upload bar stalls.
- If the file was exported from a mobile voice memo app or an unusual DAW, open it in any audio player first to confirm it plays cleanly. A file with a corrupted header can pass the size check and still fail at the parsing stage.
- If you are re-exporting to fix a format issue, export again rather than renaming the file extension. Renaming an AIFF to .wav does not change the underlying encoding and will still fail.
Song Length and Duration Requirements
The minimum song duration for Echonos Engine is 60 seconds. If your track, loop, or edit is shorter than that, the audio analysis stage has too little material to detect tempo and structure reliably, and the job fails before generation begins.
This comes up most often with three kinds of uploads: a short jingle or sting, a trimmed preview clip meant for social media rather than the full song, or an intro-only export where someone accidentally cut the file at the wrong marker. The fix is straightforward: upload the full song, or loop/extend a short piece to clear 60 seconds if you specifically need a short-form asset generated.
There is no published maximum song length, but very long tracks (10-plus minutes) take proportionally longer for the audio analysis stage to process and increase the surface area for something else on this checklist to also go wrong partway through. If you are testing the workflow for the first time, a 2 to 4 minute song is the easiest case to debug if something does fail.
One less obvious duration issue: some export tools embed a duration metadata tag that does not match the actual audio length, usually from a file that was trimmed in a waveform editor without re-encoding properly. If your file reports as long enough in your file browser but still fails duration checks, re-export from the original source rather than trying to patch the existing file.
Credits, Queues, and Retrying Safely
If the upload succeeds and the creative direction is accepted but the generation still does not complete, the issue is usually account state rather than the file.
Check your credit balance first. A full Engine generation is 200 credits flat, regardless of song length. If your balance is below 200, the job will not start, and depending on where in the flow this is checked, you may see an error immediately or after the upload completes. New accounts get 250 free signup credits, which covers exactly one full Engine generation with a small amount left over. If you have already used your signup credits on a first attempt and this is a second song, check your balance before assuming something is broken.
Check whether the job is actually stuck versus just running. A full generation takes minutes, not hours, but "minutes" during a busy period can feel longer than it is. Before assuming a failure, check the job status in your dashboard. A job showing "processing" is working. A job showing an explicit error state has actually failed and is safe to retry or investigate further.
Network interruptions mid-upload cause a specific failure mode. If your connection drops while the audio file is uploading, the job can be left in a partial state that neither completes nor cleanly errors. If a job has been sitting in an ambiguous state for longer than a normal generation should take, cancel it and start a fresh upload rather than waiting indefinitely.
Retrying safely means changing at least one variable before you submit again, or confirming the failure was environmental (network, browser tab closed, connection dropped) rather than something in your input. If you retry the exact same file and prompt after a network-related failure, that is a reasonable first retry. If you retry the exact same file and prompt after a format or duration failure without fixing the underlying file, you will get the same failure again.
If a job fails repeatedly with the same file and the file passes all the format, size, and duration checks above, that is the point to treat it as an account or platform issue worth checking on rather than continuing to retry blind.
When to Regenerate vs Start Clean
This is the decision that saves the most time and credits, and it is also the one artists get wrong most often.
Regenerate a single scene in Studio when the project mostly worked. If your video generated successfully and only one or two scenes look wrong (a shot that missed the mood, a character that looks slightly off in one frame, a transition that feels abrupt), that is not a failed generation. That is a normal part of the process, and Studio's scene-level regeneration exists specifically for this. You select the scene, adjust the prompt, and regenerate just that shot for a flat 50 credits (video regen) or 10 credits (image regen), while everything else on the timeline stays untouched. Starting a full Engine re-run to fix one weak scene is the single most common way artists waste credits on this platform.
Start a full generation clean in Engine when the problem is systemic or the upload never completed. If the job failed before producing any output, if the character does not resemble your artist across the entire video, if the art style is wrong throughout, or if the audio itself was rejected, there is nothing to salvage at the scene level, and a full Engine run (with the underlying issue fixed) is the correct move.
A simple test: count how many scenes are actually wrong. One or two, fix them in Studio. Most of them, or the generation never completed at all, go back to Engine and run it again with the corrected input.
If your video came out with a specific quality problem rather than an outright failure, the fix usually lives in Studio rather than a restart. Video that has drifted from the beat is covered in the guide to fixing an AI music video that is out of sync. Faces that shift or warp between scenes are covered in the guide to stopping morphing faces in AI video. Visible glitches or compression artifacts scattered through a generated video are covered in the scene-by-scene approach to cleaning up AI video artifacts.
Preventing Failures on Your Next Upload
A short pre-upload routine catches most of what this checklist covers, before you spend a generation attempt finding out the hard way.
- Confirm the audio file is MP3, M4A, WAV, AAC, OGG, or FLAC, and under 40 MB.
- Confirm the song is at least 60 seconds long, and that the file plays cleanly start to finish in a normal audio player.
- Check your credit balance before uploading, especially if this is not your first generation of the session.
- Keep the creative direction specific: name the world, the character, the camera language, and the lighting. A vague or contradictory prompt does not usually cause an outright failure, but it increases the odds the generation "succeeds" in a way that reads as a failure once you watch it back. The AI music video prompt guide covers prompt structure in depth.
- If you are setting up a character for the first time, upload a clean headshot at minimum, and add the optional Full Body, Left Profile, and Right Profile references if you have them. The character consistency guide covers why this matters for keeping the same face across every scene.
- If a generation from audio has failed before for a reason you could not diagnose, the guide to generating an AI music video from your own audio walks the upload-to-output path end to end and is a useful reference for where in the pipeline something can go wrong.
When to Escalate
If you have confirmed the file meets every format, size, and duration spec, confirmed your credit balance covers the generation, confirmed the job shows an explicit error state rather than just "processing," and the generation still fails, that is the point to treat it as a platform issue rather than something fixable from your end. Note the exact error message if one is shown, the approximate time the job failed, and the file details, and reach out through the in-app support channel with those specifics. A vague "it did not work" report takes longer to resolve than one with the job details attached.
FAQ: Errors, Credits, and Supported Formats
Why does my AI music video generation keep failing?
The most common causes, in order of frequency, are an unsupported audio format (AIFF is not accepted; use MP3, M4A, WAV, AAC, OGG, or FLAC), a file over the 40 MB size limit, a song under the 60-second minimum duration, or an account credits balance below the 200 credits a full Engine generation requires. Work through the format, size, and duration checks first since they account for most upload-stage failures, then check your credit balance and job status if the file itself checks out.
What audio formats does Echonos accept for video generation?
MP3, M4A, WAV, AAC, OGG, and FLAC, up to 40 MB, with a minimum song length of 60 seconds. AIFF, ALAC, WMA, Opus, and DSD are not supported. If your master is in one of those formats, export to WAV or a high-bitrate MP3 before uploading.
How many credits does a failed generation cost me?
A generation that fails before completing (an upload rejected for format, size, or duration reasons, or a job that errors out during processing) should not consume your credits. Credits are charged for a completed generation. If you believe you were charged for a failed job, that is worth flagging to support with the job details, since it is not the intended billing behavior.
Should I regenerate the whole video or just fix one scene?
If only one or two scenes look wrong and the rest of the video works, use Studio's scene-level regeneration: 10 credits for an image regen, 50 credits for a video regen, both flat fees regardless of the scene's length. If the problem affects most of the video, the character is inconsistent throughout, or the generation never produced usable output at all, run a full Engine generation again after fixing the underlying issue. Regenerating scene by scene to patch a project-wide problem usually costs more in total credits than a single clean re-run.
My video generated but looks nothing like what I described. Is that a failure?
Not a technical failure. The pipeline completed and produced output; the output just did not match your intent. This is usually a prompt issue rather than an error, and the fix is to rewrite the creative direction with more specific language (naming the world, the character, the camera movement, and the lighting rather than mood words) and regenerate the affected scenes in Studio. The AI music video prompt guide covers the language patterns that produce output closer to what you intended on the first pass.
Why did my upload succeed but the generation never started?
This usually points to a credits or queue issue rather than the file. Check your account balance first: a full Engine generation requires 200 credits. If your balance covers it, check your job status in the dashboard; a job can take several minutes to move from queued to processing during busy periods, which can look like nothing happened. If the job has been in an ambiguous state well beyond a normal generation window, cancel it and start a fresh upload rather than waiting indefinitely.
Keep reading
Related Articles
Cleaning Up Artifacts in AI Music Videos: A Scene-by-Scene Approach
AI music video artifacts like warping or smearing usually live in one or two scenes. Here is how to spot them and fix each scene without a full re-render.
Fixing Choppy AI Video: Frame Rate, Motion, and Smooth Playback
Fix ai video choppy playback with motion-aware prompting, Studio scene regeneration, and a checklist for telling a generation issue from a local one.
When AI Video Does Not Match Your Lyrics: How to Align Scenes to Meaning
Fix ai video not matching lyrics with section-by-section prompting, scene-level regeneration, and a lyric-to-timeline plan you build before you generate.
Written by
Echonos Team
We build Echonos — an AI music video pipeline for indie artists, managers, and small labels. We write here about how we think about audio, visuals, and release workflow.