Docs

The desktop app names failures with a short code. Copy for support on a queue row puts that on a code: line, plus stderr. Paste the whole text into the assistant — the Worker looks the code up. Do not attach the clip.

code: unknown means the engine exited before it emitted a named code. The assistant still classifies common stderr (CUDA OOM, disk full, missing encoder). If nothing matches, use Talk to a human.

See troubleshooting for preflight (VRAM, denoise fallback, Remaster CUDA).

Codes

cuda_oom

GPU ran out of memory. Close other GPU apps, lower tile size in Settings, or use Export Ready instead of Remaster.

disk_full

The temp or output drive is full. Free space, or point Settings → temp folder at a larger drive.

encoder_missing

FFmpeg lacks the encoder for that delivery format. Switch to H.264, or install a full FFmpeg and set its path in Settings.

corrupt_input

The source is missing, truncated, or not a format FFmpeg can decode. Confirm the path, play it elsewhere, remux to H.264 MP4 if needed.

ffmpeg_failed

An FFmpeg subprocess failed during extract, encode, or remux. Read the stderr tail. Try bundled FFmpeg and H.264 delivery.

scenedetect_unavailable

Detect Cuts needs PySceneDetect. Use manual splits, or reinstall so the engine includes it.

backend_unavailable

That AI stage has no working backend, usually a model weight that is not downloaded. Settings → Hardware → download missing weights, or turn the feature off for this run.

compare_source_missing

Before/after needs both files on disk. Restore the missing path or re-run the job.

compare_output_unwritable

Could not write the comparison MP4. Check free space and that nothing has the previous file open.

compare_font_missing

Burn-in needs the font bundled with the engine. Reinstall, or turn burn-in off if you only need the wipe.

framediff_source_missing

Review difference view is missing a before or after file. Restore the files and open Review again.

framediff_ffmpeg_failed

FFmpeg could not build the Review difference view. Confirm both files play, then retry Review.

engine_not_configured

No Python engine. Reinstall so the bundled engine is restored, then check Settings → Engine.

engine_spawn_failed

The engine binary was found but did not run. Run Test engine in Settings → Engine. Clear a bad Python override. Allow FrameLathe through antivirus / SmartScreen.

engine_argv_mismatch

App and engine are out of step. Update FrameLathe so they ship together.

vsr_torch_missing

Remaster needs PyTorch in the interpreter that is actually running. Settings → Hardware → Download GPU engine (~2 GB). Use Export Ready on Vulkan/ncnn.

vsr_cuda_missing

Remaster needs CUDA torch, not a CPU wheel. Settings → Hardware → Download GPU engine. Remaster needs an NVIDIA CUDA card.

vsr_weights_missing

RealBasicVSR checkpoints are not fetched with the other models. Settings → Developer → Check / fetch Remaster weights. Do not reinstall torch for this code.

sidecar_stale

Bundled engine does not match this build. Update FrameLathe.

trt_cuda_missing

TensorRT build needs CUDA torch. Download the GPU engine, or skip TensorRT — upscale already runs on CUDA torch.

trt_build_ram

Not enough free system memory to build TensorRT. Close a browser and retry, or skip it. Nothing is wrong with the card or the download.

trt_not_installed

TensorRT for RTX is missing from this engine. Download the GPU add-on pack again, or skip TensorRT.

hub_model_licence_blocked

This Hub or imported model is blocked because the Hub licence filter is on, and the recorded licence string looks non-commercial, restricted, or missing. Clear the custom model on the row, pick a model the filter allows, or turn the filter off in Settings → Model Hub. FrameLathe does not license Hub weights; the publisher’s terms still govern.

hub_slot_closed

This stage has no bring-your-own slot. Colorize runs first-party DDColor. Use the Upscale, Denoise, or Matting tab instead, or clear the custom model on the queue row.