Easy-dlp

A dark, liquid-glass desktop front end for the yt-dlp command line, with a batch download queue.

Run it

Double-click dist\Easy-dlp.exe. That is a single packaged file with Python, tkinter and Pillow inside it, so nothing needs installing and you can move it wherever you like (Desktop, Start menu, pinned to the taskbar).

It still calls the real yt-dlp and ffmpeg on your PATH, which is what keeps the download engine current — update yt-dlp whenever a site changes and the app picks it up. If yt-dlp is missing the app opens and says so rather than failing silently.

To run from source instead:

python app.py

To rebuild the executable after changing the code:

python build_exe.py

Requirements

Needs	Why	Install
yt-dlp on PATH	does the downloading	winget install yt-dlp.yt-dlp
ffmpeg on PATH	merging, audio extraction, embedding	winget install yt-dlp.FFmpeg
Python 3.10+ with tkinter	only to run from source or rebuild	ships with python.org installers
Pillow	only to run from source	pip install -r requirements.txt
PyInstaller	only to rebuild the .exe	pip install pyinstaller

The app finds yt-dlp on PATH; running from source it also falls back to python -m yt_dlp. If neither is there it still opens, warns you, and every job fails with a clear message. The header shows the detected yt-dlp version and whether ffmpeg was found.

Easy-dlp.bat is a convenience launcher for the source version (starts it without a console window).

Staging and downloading

The left column is a staging list, so everything happens in this one window instead of keeping a browser, a text file and a downloader open side by side.

1. Each staged row is a pair of boxes: FILE NAME on the left, URL on the right. The name is optional.
2. Add appends another empty pair. It does not download anything, so you can build the whole list first. Pressing Enter in a URL box does the same.
3. The x at the end of a row removes that row. Hold Shift while clicking it to remove every row in the same numbered section, which clears out an opening section of twenty odd intro lessons in one click.
4. Paste fills empty rows from the clipboard, one row per link, adding rows as needed; Import does the same from a .txt file; Clear empties the list.
5. Paste a whole course page instead and it stages the entire course at once — see below.
6. Start Download, at the foot of the options below, queues every row that has a URL and starts working through them. The staging list then resets, ready for the next batch. It sits last on purpose: everything that decides where a download goes — the folder above all — is read on the way down to it, rather than being scrolled past after the fact.

Staging a whole course from its page

Fishing each link out of the network tab is the slow part, so the app will read a course page instead:

1. Open the course page in the browser, where every lesson is listed.
2. Ctrl+U to view the source, then Ctrl+A, Ctrl+C.
3. Ctrl+V in the app, or press Paste.

Every lesson is staged, named Section.Lesson_Title — 1.18_CreateABlurredBackground — so they land in the folder in course order without renaming anything. Saving the page with Ctrl+S and using Import works too.

This reads the video id out of each lesson's own thumbnail, which is how Bunny Stream (b-cdn.net) publishes it; it is not a general scraper. On a site that hosts video another way nothing is found and the paste falls back to being read as a plain list of links — see Which players work how below.

Course resources

A course is rarely only its videos. Worksheets, brush sets, project files and reference sheets come with it, and those are staged too: when a page offers a handout, it is added after the lessons and downloaded as the file it is.

The rule is deliberately narrow, so a paste does not drag in every image on the page. A link ending .pdf, .zip, .psd, .brushset and the like is a handout by any reading, and comes along. A link ending .png or .jpg is as likely to be the site's logo or somebody's avatar, so an image is only taken when the page says outright that it is meant to be saved — a download attribute, or wording that says so.

The two collector scripts do the same while they walk a course, and there they can do better: a handout keeps the number of the lesson it belongs to, so 2.05_ColourTheory_Worksheet.pdf sorts beside 2.05_ColourTheory.mp4. Anything the course page itself offers, rather than one lesson, is numbered 00_.

Turn it off with the Resources switch under Download options if you only want the videos.

Rows without a URL are ignored, and anything that is not an http(s) link is skipped with a note rather than queued.

The staging list is saved as you build it, so closing the app — or losing it — does not cost you the batch: it is still there the next time you open the window. It is only cleared once you press Start Download.

Parallel downloads (1–6) decides how many run at once. Set it to 1 and the queue works strictly one at a time; at 2 or more, that many run together and the rest wait their turn.

Each queue row shows live percentage, speed, ETA and transferred size, and while a playlist is running it shows 4/13 for the item it is on.

Options are captured per job when you press Start Download, so changing them afterwards affects only what you stage next.

Which players work how

Course sites hand their video to one of a handful of players, and which one decides how much work a batch takes. What follows was tested against a real course on each.

Player	Tell-tale in the address	What you paste	Sign-in	Whole course at once
Bunny Stream	b-cdn.net	the course page source	no	yes — paste the page
Cloudflare Stream	cloudflarestream.com	signed links, ~6 hours	carried in the link	yes — console tool
Wistia	wistia in the page	the lesson page address	cookies.txt	partly — links are listed
Vimeo	player.vimeo.com/video/…	that address	usually none	no — one at a time

Bunny Stream — paste the page

The video id alone plays, with nothing to authorise it, so the whole course comes out of one page. See Staging a whole course from its page above. Tested on a 53 lesson course: all 53, correctly numbered.

Cloudflare Stream — run the collector

The video id on its own is refused with a 401 here. Playback is authorised by a signed token that lasts about six hours and is never in the page source, so pasting the page achieves nothing — the links have to be caught as each lesson plays.

1. Open the course, signed in.
2. Press play on the first lesson and let it start. Browsers block playback until you have interacted with the page, and with no playback there is no link to catch.
3. F12 → Console → paste the whole of tools/cloudflare-stream-links.js → Enter.
4. Wait, roughly ten seconds a lesson. A panel appears with the links.
5. Copy, then Paste in the app.

Tested on a six lesson course: 6 of 6, named 01_Introduction and so on. Download them the same day, because the links expire.

For one lesson without the tool: play it, F12 → Network → filter m3u8 → copy the …/manifest/video.m3u8 address.

There is also a single-file …/downloads/default.mp4 address on these videos. yt-dlp cannot use it — it recognises the Cloudflare domain and runs its own extractor regardless, which then fails with 401.

Wistia — cookies, then lesson addresses

The best downloads of the lot: up to 4K, delivered as one plain file rather than hundreds of fragments, which is far kinder to a network drive. These sites usually sit behind a sign-in.

1. Export a cookies.txt for the site with a cookie extension.
2. Set COOKIES to cookies.txt file and choose it.
3. Paste lesson page addresses — the ordinary ones from the address bar. There is no need to find the video address at all.
4. Put something in the FILE NAME box: left alone, the file is named after an internal upload path.

For a whole course, the course page lists every lesson address; copy them out and paste the lot. Do not paste the raw page source here — unlike Bunny, that pulls in every link on the page.

Vimeo — run the collector, or copy one address

For a whole course, tools/vimeo-course-links.js collects the lot:

1. Open the course page — the one listing the lessons — while signed in.
2. F12 → Console → paste the script → Enter.
3. Wait a few seconds a lesson, then Copy and Paste in the app.

Lesson addresses are listed on the course page, but each lesson's Vimeo id is only put into its own page by scripts after it loads, so fetching those pages in the background finds nothing. The script gets around that by loading each lesson in a hidden frame, which runs its scripts as normal, then reading the player address out of it. Nothing is downloaded and nothing on the site is changed. Tested on a seven lesson course: 6 collected, the other having no video on it.

For a single lesson, skip the script: right-click the video → Copy video address, or find the <iframe> in F12 → Elements. You want https://player.vimeo.com/video/….

Course-embedded Vimeo videos often need no cookies and no referer at all.

Running the collector scripts without pasting them every time

A .js file cannot be dragged into the Console — it has to be text. Pasting it works (the first time, Chrome makes you type allow pasting before it accepts one), but for anything you will do twice, keep it as a snippet:

1. F12 → Sources → Snippets in the left pane (under » if the pane is narrow).
2. New snippet, paste the script in, give it a name.
3. On any course page after that: open the snippet and press Ctrl+Enter.

Snippets run against whatever page is in front, so one copy serves every course on that site, and they survive restarts.

Signing in

"Cookies from browser → chrome" no longer works. Recent Chrome encrypts its cookie store, so yt-dlp reports Failed to decrypt with DPAPI and the site redirects to its login page. Export a cookies.txt with a browser extension and pick cookies.txt file, which is the first entry in the COOKIES list and takes precedence over the browser setting. Cookies expire, so re-export when a download suddenly lands on a login page.

Sites behind Cloudflare's bot check refuse a plain request with a 403. The app passes the argument that gets around it on every run; it is aimed at the generic extractor, so nothing changes for a direct media link.

When something fails

What you see	What it means
redirected to a login page	cookies missing or stale — export again
HTTP 403 immediately	the site's bot check; already handled, so the site has tightened up
HTTP 401 on a signed link	the token expired — collect the links again
Failed to decrypt with DPAPI	the Chrome cookie problem above
dies part way to a network drive	usually the share dropping out; check Activity and keep Parallel low

Nothing is lost

Three things are kept on disk so a closed window, a crash or a failed batch never costs you a list of links you had to go and collect:

• The staging list is saved as you type it, and comes back next time.
• The queue is saved too. Close the app mid-batch and the items are still there when it reopens, with their file names, their status and their progress. Nothing restarts by itself, so press Start when you are ready.
• Every batch you send to the queue is written out as a plain list, both into the download folder as _download-list_<date>_<time>.txt and into a history folder beside the settings. Each line is url | file name, which is exactly what Paste and Import read, so recovering a batch is copy, paste, Start Download.

The Batches button in the Activity window opens that history folder; the last hundred batches are kept.

Naming the output file

By default files are named by the filename template under Advanced options. There are two ways to override that for one video:

Before it is queued — type into the FILE NAME box next to its URL.

After it is queued — hover the row in the queue and press the pencil. That opens a small sheet for the output name. Renaming is offered for items that are waiting, on hold, failed or stopped; it is deliberately not offered while an item is downloading, paused or finished, because a partly-downloaded .part file (or a file already on disk) is tied to its current name.

You type the name, not the extension — that is added for you, and .mp4 typed on the end is dropped rather than doubled. Characters Windows forbids are removed, / becomes -, and the name is capped at 150 characters. yt-dlp fields still work, so %(title)s (2026) is a valid name. If a named link turns out to be a playlist, the index is appended (Name - 001.mp4) so the videos cannot collapse onto one file.

A .txt file being imported may still name its links with a pipe (https://... | My Name), which is handy for lists produced elsewhere.

Downloading to a network drive

Downloads are built in a private local scratch folder and only the finished file is moved to the destination. Nothing partial is ever written to the destination, which matters for a NAS or any share reached over a VPN or Tailscale: yt-dlp writes each HLS fragment as its own small file, so a ten minute video means hundreds of small writes, and one truncated write turns into a decryption error ("Data must be padded to 16 byte boundary in CBC mode") that looks like a corrupt video but is not.

Each job gets its own scratch folder, so parallel downloads can never share a .part file even when two videos resolve to the same title. Scratch folders are removed when a download finishes, and stale ones are swept after a week.

A failure that looks temporary — a share that blinked out, a refused key, a truncated fragment, a 5xx or 429 — is retried automatically twice before the item is marked Failed, and the retry asks for one fragment at a time with spacing between requests rather than hammering the same host again. Permanent failures (a dead link, a private video, an unsupported site) are reported immediately instead of being retried pointlessly.

Every line of yt-dlp output is also written to activity.log next to the settings, so a failure can be diagnosed after the fact even if the window was closed.

Pausing and stopping

Pausing is real: the download stops, the partly-downloaded .part file stays, and resuming carries on from where it left off instead of starting again.

One item — hover a row. A tick box appears on the left and the actions for that row appear on the right: pause (or resume), stop, and remove. A finished row offers reveal-in-folder and download-again instead.

Several items — tick the rows you mean. That arms the Pause and Stop buttons in the queue toolbar, which are greyed out until something is ticked. Pause becomes Resume when everything ticked is already paused. The tick box next to the QUEUE heading selects every row.

Everything — the primary button on the right is the whole-queue control: Start works the queue and releases anything paused or held, and it turns into Pause all while downloads are running.

Two states mean "not running, will continue later":

• Paused — it was downloading, and has partial data to resume from.
• On hold — it never started. Pausing a waiting item holds it, so you can park something you added and let the rest of the queue go first, then tick it and press Resume when you want it.

Options

Quality — Best available, 2160p/1440p/1080p/720p/480p (each capped at that height, falling back down when nothing matches), or Audio only. Container — Auto, mp4, mkv or webm. With mp4 the picker prefers mp4+m4a streams so no re-encode is needed. Audio format — mp3, m4a, opus, flac or wav, at best quality (enabled only for Audio only).

Switches: Subtitles (download and embed, languages configurable), Thumbnail (embed as cover art), Metadata (title, artist and chapters), SponsorBlock (cut sponsor/self-promo/interaction segments), Playlists (follow a playlist link instead of grabbing one video), Skip existing (keeps .evd-archive.txt in the download folder and skips anything already in it), Resources (stage the handouts a course page offers alongside its videos).

A handout skips the video machinery entirely: no format picking, no container merge, no tag embedding — those would either do nothing to a PDF or corrupt it. It is fetched under the name it was staged with, extension and all. The speed limit still applies, since it is the same connection.

Under Advanced options: speed limit (e.g. 4M), cookies from an installed browser (for private or age-gated content), subtitle languages, and the yt-dlp output template.

Playlist URLs are filed into <Playlist title>/001 - <title>.<ext> subfolders. Detection is host-aware — it looks for list=, /playlist, YouTube channel forms, SoundCloud sets, Bandcamp albums and Vimeo channels — so a normal video never gets filed into a playlist folder.

Activity opens the raw yt-dlp output for every job, which is the first place to look when a site misbehaves.

Settings are saved to %APPDATA%\EasyVideoDownloader\settings.json.

How the interface is drawn

Tk widgets cannot be translucent, so the UI is not built from them. The whole window is a canvas, and every surface is an RGBA image composited by Pillow:

• graphics.py renders the aurora backdrop, then samples and blurs it per panel to produce genuine frosted glass with a rim light, sheen and shadow — Tk 8.6 alpha-composites canvas images over each other, so translucent surfaces really show what is behind them.
• widgets.py is a small canvas toolkit (buttons, selects, switches, fields, progress bars, tooltips). Only text entry uses a real Tk widget, since a caret cannot be faked; those are colour-matched to the glass behind them.
• The left column and the queue each live on a nested canvas so they clip and scroll independently.

Two consequences worth knowing:

• The layout is expressed in design pixels and Tk's point-to-pixel conversion is pinned, so text and layout always agree. The window is DPI-unaware, which means Windows scales it to match your display scaling; on a high-DPI display it is very slightly softer than a natively scaled app.
• Resizing regenerates the backdrop and panels, so it is debounced.

Layout

app.py                  entry point
evd/theme.py            colours, type scale, metrics
evd/graphics.py         Pillow compositor: backdrop, frost, icons
evd/widgets.py          canvas widget toolkit
evd/queueview.py        scrolling queue with pooled rows
evd/downloader.py       yt-dlp process management and progress parsing
evd/config.py           settings persistence
evd/errors.py           crash logging
evd/ui.py               window layout and wiring
build_exe.py            packages it into dist/Easy-dlp.exe

Progress is read back through --progress-template, which emits machine-readable fields instead of the human progress bar, so parsing does not break when yt-dlp changes its display. Final paths come from --print after_move:filepath. Cancelling kills the whole process tree, so ffmpeg children do not survive.

Notes

• Pausing and resuming is tested end to end: a job paused at 8% keeps its .part file and resumes from 8% rather than restarting.
• Anything unexpected is appended to crash.log next to the settings, because a packaged window has nowhere to print a traceback. A layout that fails says so on screen and retries rather than leaving an empty window, and the staged links survive it.
• Custom names are tested end to end too, including through the packaged .exe: My Custom Clip: Take 1? lands on disk as My Custom Clip Take 1.mp4.
• The first launch of the packaged .exe takes a few seconds while it unpacks itself to a temp folder; later launches are quicker.
• Embedding subtitles also leaves the sidecar .vtt/.srt files next to the video.
• Only download what you have the rights to, and respect each site's terms.