🎬 Floyo Video Studio

One node to turn a video into a ready-to-use VIDEO + frames + audio + fps + frame count — with time-based trim, 1-click downscale, and exact frame targeting. For ComfyUI.

TL;DR — Upload a clip and get a real VIDEO (plugs straight into video‑to‑video AI nodes) plus its frames, audio, fps and frame_count from one node. Trim by seconds (slider shows timestamp and frame number), downscale in one click, hit an exact frame count, stay memory‑light on long/4K clips, and never expose a file path.

---

✨ Features

• 🎞️ Real VIDEO output — hands downstream nodes a native ComfyUI VIDEO (file‑backed), so it connects directly to Floyo's video‑to‑video AI nodes, SaveVideo, AudioVideoCombine, etc. — not just raw frames.
• ⏱️ Trim by time, not frame index — drag a range slider in seconds; it shows the timestamp + the frame number live as you scrub.
• 🎯 Exact frame targeting — set target_frames to force a precise count (e.g. a model that needs a multiple of 16); fps is derived automatically. Or set target_fps instead. (target_frames wins when both are set — industry‑standard: frames = fps × duration.)
• 📉 1‑click downscale — Source / 1080p / 720p / 480p. Never upscales.
• 🔎 Instant source info — read‑only FPS / Frames of the loaded clip, read from the file header server‑side (no decoding) — so it stays fast and light even for 20–30 min 4K uploads on low‑end devices.
• 🔊 Audio passthrough — outputs the trimmed audio (and muxes it into the VIDEO); wire it to any audio input.
• 🪶 Memory‑light — seeks, scales and drops frames inside the decode, so a 3s slice of a long 4K clip never decodes the whole file at full res. No OOM.
• 🔒 Upload‑only, no file paths — resolves a filename strictly inside ComfyUI's input dir (traversal‑guarded); nothing path‑like is returned or saved into the workflow. Built for hosted ComfyUI.

---

📤 Outputs

Port	Type	What
video	VIDEO	The trimmed/scaled clip as a native, file‑backed video — for AI video nodes & SaveVideo.
frames	IMAGE	The trimmed/scaled frame batch.
audio	AUDIO	The trimmed audio (silent if the source has none or include_audio is off).
fps	FLOAT	Output fps.
frame_count	INT	Number of output frames.

🎛️ Inputs

Widget	Type	Does
video	upload	Pick / upload a clip — filename only, no paths. Shows an in‑node preview.
start_seconds	float	Trim start (seconds).
end_seconds	float	Trim end (seconds). 0 = end of clip.
quality	preset	Downscale: Source / 1080p / 720p / 480p. Never upscales.
target_fps	float	Output fps. 0 = keep source.
target_frames	int	Exact output frame count. 0 = off. Overrides target_fps.
include_audio	bool	Also output the trimmed audio.

The panel also shows a live Output readout (frames · fps · duration) and greys out target_fps when target_frames is set, so the active mode is always clear.

---

📦 Install

ComfyUI Manager — search Floyo Video Studio → Install.

Git

cd /path/to/ComfyUI/custom_nodes
git clone https://github.com/FloyoAI/Floyo-Video-Studio.git
pip install -r Floyo-Video-Studio/requirements.txt   # PyAV

Restart ComfyUI → Add Node → Floyo → Video → 🎬 Floyo Video Studio.

Needs PyAV (av) — it bundles ffmpeg, so there's no separate binary to install.

---

🚀 Quick start

1. Add the node and upload a video.
2. Drag the trim slider to the range you want (watch the timestamp + frame number).
3. Optionally pick a quality preset and/or set target_fps or target_frames.
4. Wire it up:
   • video → any video‑to‑video AI node or SaveVideo
   • frames → image nodes · audio → audio nodes · fps / frame_count → wherever a count is needed.

Example (proven end‑to‑end): Floyo Video Studio.video → Seedvr_Upscale_Video → Video URL to Frames → Video Combine.

---

📝 Notes

• Why a VIDEO output: every Floyo video‑to‑video AI node consumes the native VIDEO type. The output is encoded to a temp file so partner uploaders that need a real path work out of the box.
• target_frames vs target_fps: for a fixed trim duration, an exact frame count forces fps = frames / duration — matching FFmpeg / NLE retiming / AI video models. (A few fps‑conditioned models, e.g. SVD, treat fps as a motion input — set target_fps there instead.)
• Hosted‑ComfyUI safe: no path in, no path out; streaming decode keeps memory flat on long/4K clips.

---

📄 License

MIT — see LICENSE.

Built with care by Floyo 💜