The actual output files
These are the files those runs wrote, unedited, and they are included in the download.
Every workflow in this pack was submitted to a live ComfyUI instance and had to return
a real output file before it was allowed in the box. The images were then checked for
pixel variance — a 1.4 MB PNG can still be a blank frame — and every video was inspected
with ffprobe. The files those runs produced ship with the pack, so you can
compare against your own before you spend a byte on model weights.
Runtimes are wall-clock from ComfyUI's own execution log on one RTX 5090 (32 GB), ComfyUI 0.21.1. Yours will differ with VRAM, precision, and disk speed. These are smoke-test outputs with prompts left as-is, not curated marketing renders.
| # | Capability | Nodes | Models | Runtime | Verified output |
|---|---|---|---|---|---|
| 1 | Text to image, Flux | 8 | 1 | 38.0 s | 1024×1024 PNG · 1.48 MB |
| 2 | Image-to-image edit, Flux Kontext | 13 | 4 | 25.5 s | 1024×1024 PNG · 1.90 MB |
| 3 | Stylised manga frame | 7 | 1 | 3.6 s | 768×512 PNG · 548 KB |
| 4 | Image to video, Wan 2.2 + Lightx2v | 17 | 6 | 52.0 s | 640×640 h264 · 49 frames · 3.06 s |
| 5 | Text to video with audio, LTX 2.3 | 29 | 4 | 107.4 s | 1536×1024 h264 + AAC · 97 frames · 3.88 s |
| 6 | Image to video with audio, LTX 2.3 | 35 | 4 | 70.4 s | 1536×1024 h264 + AAC · 97 frames · 3.88 s |
| 7 | 3× interpolation + 4× upscale, RIFE + RealESRGAN | 6 | 2 | 145.9 s | 145 frames @ 1280×1280 |
These are the files those runs wrote, unedited, and they are included in the download.
18 distinct model files. They are third-party open weights, so redistributing them here would be both unlawful and a 100 GB download you did not ask for. The manifest names every file with its exact size on disk, which is the point — downloading the wrong variant is the expensive mistake this pack exists to prevent.
Everything was verified on one RTX 5090 (32 GB). The image workflows are modest; the video ones are the VRAM-hungry pair. On 24 GB expect to drop resolution or frame count. Below that, the video workflows are not realistic.
It reads frames from output/wan_frames, which an upstream workflow must
write first. Run it cold and it fails with "Folder cannot be found" — that is expected,
not a defect. The README documents the exact chain that was verified, including the
ffmpeg command.
LoadImagesFromFolderKJ (KJNodes) and RIFE VFI
(ComfyUI-Frame-Interpolation). Both free, both listed in the README with what needs them.
docs/verify-yourself.py is the same script used to produce the runtime table
above. Point it at your own ComfyUI and it will submit each workflow and report which ones
returned a file. If our numbers are wrong on your hardware, you will find out in one command.
# submit-only: catches a missing model in seconds python docs/verify-yourself.py --validate # full run: waits for outputs, prints what each workflow wrote python docs/verify-yourself.py --run
Stripe-hosted checkout — card details never touch an ABUZ8 server. After payment you are
sent a signed download link for a 6.4 MB zip
(SHA256 f0a6e3e7…6e5f3571, published so you can
verify the file you receive is the file we shipped).
If a workflow does not run, if the manifest is wrong, or if the download fails,
email [email protected] and we refund it. We do not
claim "tested before it's sold" as a substitute for that — we tell you exactly what was
tested, on what, and when, and you hold us to it.