VELVET VICE LTX — v0.1.10 Fast NVENC Output + Preview
The v0.1.10 video output — same single encode, but it hands you the MP4 for a preview
- images
- audio
- Filenames
VelvetViceLTXAutoVideoCombineV0110 is the version-bound final-output node pinned by the v0.1.10 Velvet Vice LTX workflow. If you're running that workflow, this is the node doing your MP4 encode, and the two things it added over the earlier v0.1.9 variant were both about the preview experience, not the encode itself.
What v0.1.10 added
First, the fast FP16→RGB8 frame delivery. When your frames arrive as FP16 - which is exactly what the pack's FP16 VAE Decode produces - VideoHelperSuite's normal conversion path multiplies the FP16 numpy array by 255, and on desktop CPUs FP16 arithmetic is software-emulated and becomes the dominant cost for multi-gigapixel clips. This variant (like all current Velvet Vice output nodes) swaps in a converter that promotes to FP32 first, giving mathematically correct 8-bit rounding much faster.
Second, the direct preview: the saved MP4 is returned so you get a preview of the actual encode, without a second, separate preview encode of the same frames. One encode, one result, preview for free. That's the "Fast" in the name.
Everything else matches the base Auto NVENC output node: AUTO mode probes NVENC with a 256×256 test frame and falls back to CPU H.264 at your CRF; NVIDIA NVENC ONLY raises if the probe fails; CPU H.264 ONLY skips hardware entirely; bitrate is scaled from nvenc_bitrate_mbps_at_24fps by actual FPS; and the output path is Windows-safe with %date:...% token support.
Inputs and outputs
Identical surface to the base node: images, frame_rate, loop_count, filename_prefix (video/LTX23-FINAL), encoder_mode, nvenc_bitrate_mbps_at_24fps (60), cpu_crf (15), pix_fmt (yuv420p), pingpong, save_metadata, trim_to_audio, save_output, optional audio. Output is Filenames (VHS_FILENAMES), an output node.
Installing it
It ships in VELVET VICE - LTX (velvet-vice-ltx) - ComfyUI Manager, or:
cd ComfyUI/custom_nodes
git clone https://github.com/Velvet-Vice/velvet-vice-ltx
Restart and hard-refresh with Ctrl+F5. The pack declares no pip dependencies; this node needs VideoHelperSuite and FFmpeg at runtime.
Which one to pick
If your workflow is v0.1.10, keep this node - it's pinned for a reason, and version-mixing the custom-node folder is the one thing the README forbids. Building fresh? The current default is v0.1.12 (which keeps the fast delivery and pairs it with a more stable preview widget behavior), so prefer that or the unversioned base node over a pinned legacy variant.
Troubleshooting
- "VideoHelperSuite is required" - install VHS via Manager, restart.
- Preview doesn't appear - the MP4 preview depends on the saved file being returnable; make sure
save_outputis on (default) and your output directory is writable. - NVENC errors in NVENC ONLY - that's the probe doing its job; switch to AUTO for the fallback path.
A version-bound utility node isn't glamorous, but it's the reason old Velvet Vice workflows keep working after updates: the encode behavior is locked to the version that shipped with the workflow.
Inputs (13)
| Name | Type | Default | Description |
|---|---|---|---|
| images | IMAGE | — | |
| frame_rate | FLOAT | 241–240 | — |
| loop_count | INT | 00–100 | — |
| filename_prefix | STRING | video/LTX23-FINAL | — |
| encoder_mode | COMBO | AUTO — NVIDIA NVENC / CPU H.264 FALLBACK | 3 options: AUTO — NVIDIA NVENC / CPU H.264 FALLBACK, NVIDIA NVENC ONLY, CPU H.264 ONLY |
| nvenc_bitrate_mbps_at_24fps | INT | 601–999 | — |
| cpu_crf | INT | 150–100 | — |
| pix_fmt | COMBO | yuv420p | 1 options: yuv420p |
| pingpong | BOOLEAN | false | — |
| save_metadata | BOOLEAN | false | — |
| trim_to_audio | BOOLEAN | false | — |
| save_output | BOOLEAN | true | — |
| audioopt | AUDIO | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| Filenames | VHS_FILENAMES | — |