Ragha Preview Video

Ragha Preview Video

A video preview player for ComfyUI that lives on the node: sound, A/B compare, 4K-ready, nothing saved to disk. Connect a batch of frames or a video, press Run, and a small looping preview plays right on the node, with its sound. Connect a second clip and compare the two with a drag slider or side by side.

Thank you, Kijai

This node was inspired by the Fast Preview node in ComfyUI-KJNodes by kijai: his idea that a preview does not need to be written to disk is what started this project. Ragha Preview Video is its own implementation, built for video (streaming decode, an in-memory clip shelf, its own player with sound and A/B compare). If you like it, please go and star KJNodes too.

What it does

• Plays a looping preview directly on the node, with play/seek controls.
• Works with an images batch (e.g. VAE Decode output) or a video input (e.g. Load Video).
• Sound: plays the audio input, or the video's own audio track. It starts muted (browsers only autoplay muted video); click the speaker to listen.
• A/B compare: connect images_b or video_b and the two clips play frame-aligned, with a drag slider or side by side. Good for before/after checks of upscaling, interpolation, colour grading or a new seed.
• Passthrough: images, video and audio outputs let you put the node in the middle of a workflow.
• 4K-ready: video is streamed one frame at a time (decode, shrink, encode), so a 1-minute 4K video uses about 200 MB of RAM.
• Nothing on disk: clips are kept in memory only, never in your output or temp folders.
• Cached runs are free: when nothing changed, ComfyUI reuses the result and the node shows the last preview again without re-encoding.

Install

Option 1 — ComfyUI-Manager / Comfy Registry Search for Ragha Preview Video in the Manager, or run comfy node install comfyui-ragha-preview-video.

Option 2 — git

cd ComfyUI/custom_nodes
git clone https://github.com/VRH7733/ComfyUI-RaghaPreviewVideo

Option 3 — manual Download the repo as a ZIP, extract it, and put the folder into ComfyUI/custom_nodes/. The folder must directly contain __init__.py (watch out for a doubled folder after extracting).

Then restart ComfyUI and hard-refresh the browser (Ctrl+F5).

Use it

1. Double-click the canvas, search Ragha Preview Video (category video/ragha), and add it. Its internal name is RaghaPreviewVideo. Coming from 1.0.0? See the changelog.
2. Connect images (for example from VAE Decode) or video (for example from Load Video). Optionally connect images_b / video_b to compare, and audio for sound.
3. Run the workflow. The preview appears on the node. In compare mode, drag across the picture to move the slider.

Example workflows are in examples/:

• ragha_preview_images.json — Empty Image → Ragha Preview Video (no model or GPU needed)
• ragha_preview_video.json — Load Video → Ragha Preview Video, with the video's sound (pick your own clip in Load Video)
• ragha_preview_compare.json — A/B compare: a video against an inverted copy of itself

Inputs

| Input | Meaning | |---|---| | images | A: frame batch (IMAGE). Optional if video is connected. | | video | A: VIDEO input. It keeps its own fps and trim. Used only when images is empty. | | images_b / video_b | B: a second clip to compare against A. A B frame batch plays at A's frame rate. | | audio | AUDIO to play with A. Without it, the A video's own audio track is used. | | fps | Playback speed for images. | | max_size | Longest side of the preview in pixels. Smaller is faster. | | quality | 1–100. | | max_frames | 0 keeps every frame; otherwise frames are evenly sub-sampled to this count (the duration stays the same). | | compare | slider or side by side, used when B is connected. | | include_audio | Put the sound in the preview. | | refresh_every_run | Off by default. On: re-encode on every Run even when nothing changed (the nodes after it then re-run too). |

Outputs

| Output | Meaning | |---|---| | images | The images input, unchanged. | | video | The video input, unchanged. | | audio | The audio input, or the A video's audio track. |

An output is empty when its input is not connected (for example images when you only connected video).

Requirements and performance guide

Software

• ComfyUI with the current frontend (tested: ComfyUI 0.43.1, frontend 1.57.4). No extra pip packages: it only uses torch, numpy and av (PyAV), which ComfyUI already installs.
• Works on CPU-only machines. A GPU is not needed for the preview itself.

Browser (the preview plays in the ComfyUI tab) | Browser | Status | |---|---| | Chrome / Edge (Windows, macOS, Linux) | Recommended. Frame-exact A/B sync. | | Firefox | Works. Plays H.264 using the system's codecs. | | Safari (macOS) | Should work (H.264 is built in); not tested. |

• Previews start muted because browsers only autoplay muted video. Click the speaker to hear the sound.
• Keep the ComfyUI tab in front while comparing. Browsers slow down hidden tabs, and A/B sync may drift until you come back.

RAM and speed, measured on a low-end laptop (AMD Ryzen 3 3250U, 2 cores, 6 GB RAM, CPU only) with 4K 30 fps phone-like videos (~67 Mbit/s, with sound), default settings (max_size 512, max_frames 0):

| Input video | Preview built in | Extra RAM used by the node | Preview size | |---|---|---|---| | 4K, 15 s | 9 s | ~180 MB | 0.5 MB | | 4K, 30 s | 16–22 s | ~185 MB | 1.1 MB | | 4K, 60 s | 29–35 s | ~90–195 MB | 2.2 MB | | 4K, 30 s, A/B compare (two videos) | 27–33 s | ~175 MB | 1.7 MB | | 720p, 81-frame IMAGE batch (e.g. a generated clip) | 2.6 s | ~30 MB on top of the batch itself | 0.01 MB |

(Ranges are from repeated runs; the laptop was also running a browser playing the previews.)

• RAM stays about the same whatever the video length: frames are decoded, shrunk and encoded one at a time.
• Finished clips are kept in memory (up to ~192 MB in total, oldest dropped first; each node keeps only its latest preview).
• Time is mostly spent decoding the 4K source, so it grows with video length and is faster on CPUs with more cores. A modern 6–8-core CPU should be roughly 2–4× faster than the numbers above.
• Minimum for 1-minute 4K: 8 GB system RAM (ComfyUI itself plus the node). 16 GB is comfortable alongside models.

Big video files: ComfyUI's 100 MB upload limit ComfyUI refuses uploads over 100 MB by default, and a 15 s 4K clip is already ~130 MB (1 minute is ~350–500 MB). This is a ComfyUI setting and affects every video node. Either:

• copy the video into ComfyUI's input folder and pick it in Load Video (refresh the list if it does not show), or
• start ComfyUI with a bigger limit, for example --max-upload-size 1024 (ComfyUI Desktop: Settings → Server Config).

Settings for long or 4K videos | You want | Set | |---|---| | Faster previews | max_size 256 or 384 | | Lighter previews of long videos | max_frames 240 (keeps the full duration, fewer frames) | | Smaller previews | quality 30–40 |

Compared with kijai's Fast Preview

| | Fast Preview (KJNodes) | Ragha Preview Video | |---|---|---| | Writes files to output or temp | No | No | | What you see | One still image (the first frame) | The whole clip playing, with sound | | Input | IMAGE | IMAGE or VIDEO (streamed, so 4K is fine) | | How the preview reaches the browser | Binary websocket image message | Kept in server memory; the player fetches it by a one-off token | | Run again with nothing changed | No new preview | Shows the last preview again (no re-encode) | | Compare two results | No | A/B slider or side by side | | Outputs | None | images, video, audio passthrough |

How it works

1. Stream: frames come from the VIDEO file (decoded with PyAV, honouring its trim) or from the IMAGE batch, are shrunk and encoded to H.264 one at a time (with AAC sound written alongside; WebM without sound if H.264 is unavailable).
2. Shelf: the finished clips are kept in a small in-memory store under random tokens. Nothing touches the disk.
3. Play: the node's UI output carries only the tokens. The player (web/ragha_preview_video.js) fetches each clip from /ragha_preview_video/clip/<token> and plays it on the node, keeping B frame-aligned with A.

Good to know

• Clips live in ComfyUI's memory until newer previews replace them, and are gone after a restart. Reloading the browser page clears the player; run the node again to bring a preview back.
• A VIDEO input's trim (start and duration, e.g. from a trim node) is respected, for both picture and sound.
• Hosted/cloud ComfyUI (Comfy Cloud, RunningHub, and similar) only offers nodes that the provider has installed.
• Tested with ComfyUI 0.43.1 and frontend 1.57.4. Newer or older frontends may behave differently — please open an issue.

Troubleshooting

| Symptom | Fix | |---|---| | Node missing from the list | The folder is nested one level too deep, or ComfyUI was not restarted. Check the startup log's Import times for custom nodes. | | "This preview is no longer in memory" | ComfyUI was restarted or many newer previews pushed it out. Run the node again. | | Blank player | Check the terminal and the browser console (F12) for lines containing Ragha Preview Video. | | No sound | Click the speaker button. If it is missing, the clip has no audio: connect audio or use a video with sound, and keep include_audio on. | | No module named 'av' | pip install av in ComfyUI's Python environment. | | Preview takes long on a 4K video | Lower max_size (256) or set max_frames (240). See the guide above. | | A and B drift apart | Keep the ComfyUI tab in front; browsers slow down hidden tabs. |

Licence

GPL-3.0. See LICENSE.

Acknowledgements

• kijai — ComfyUI-KJNodes: his Fast Preview inspired this node.
• The ComfyUI team — for the open, extensible server and frontend this builds on.