Sora 2 Videos Profile/Drafts Downloader

Downloads media URLs discovered from:

• https://sora.chatgpt.com/login (opened first by default so you can sign in)
• https://sora.chatgpt.com/profile
• https://sora.chatgpt.com/drafts

Use this only for content you own or are authorized to download.

1) Setup

python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install -r requirements.txt
python -m playwright install chromium

capture defaults to your installed Google Chrome (--browser-channel chrome). If you do not have Chrome, install it or run with --browser-channel msedge or --browser-channel chromium.

CLI usage (live)

Generated from:

python sora_downloader.py --help
python sora_downloader.py capture --help
python sora_downloader.py download --help

Top-level:

usage: sora_downloader.py [-h] [--output-dir OUTPUT_DIR] {capture,download} ...

Capture:

usage: sora_downloader.py capture [-h] [--targets [TARGETS ...]]
                                  [--seconds-per-target SECONDS_PER_TARGET]
                                  [--profile-dir PROFILE_DIR]
                                  [--login-timeout-ms LOGIN_TIMEOUT_MS]
                                  [--nav-max-attempts NAV_MAX_ATTEMPTS]
                                  [--nav-retry-delay-ms NAV_RETRY_DELAY_MS]
                                  [--goto-wait-until {commit,domcontentloaded,load,networkidle}]
                                  [--browser-channel {chromium,chrome,msedge}]
                                  [--bootstrap-url BOOTSTRAP_URL]
                                  [--no-bootstrap] [--headless]
                                  [--verification-prompt]
                                  [--no-auto-explore]
                                  [--auto-scroll-steps AUTO_SCROLL_STEPS]
                                  [--auto-step-wait-ms AUTO_STEP_WAIT_MS]
                                  [--auto-click-interval AUTO_CLICK_INTERVAL]
                                  [--auto-max-seconds AUTO_MAX_SECONDS]
                                  [--no-detail-download-links]
                                  [--passive-media-capture]
                                  [--detail-dom-fallback]
                                  [--max-detail-pages MAX_DETAIL_PAGES]
                                  [--detail-gather-max-rounds DETAIL_GATHER_MAX_ROUNDS]
                                  [--detail-gather-idle-rounds DETAIL_GATHER_IDLE_ROUNDS]
                                  [--detail-gather-pause-ms DETAIL_GATHER_PAUSE_MS]
                                  [--detail-settle-ms DETAIL_SETTLE_MS]
                                  [--detail-download-timeout-ms DETAIL_DOWNLOAD_TIMEOUT_MS]

Download:

usage: sora_downloader.py download [-h] [--manifest MANIFEST]
                                   [--storage-state STORAGE_STATE]
                                   [--overwrite] [--upload-dir DIR]
                                   [--timeout-ms TIMEOUT_MS]

2) Capture media URLs

Capture opens a visible browser (default: installed Chrome) with a persistent profile. It loads https://sora.chatgpt.com/login once first so you can sign in, then opens each target. After navigation it prints [nav] url=... title=... so you can confirm you reached the real site.

Unattended by default: capture does not wait for Enter and does not add idle time after each target (--seconds-per-target defaults to 0). Use a --profile-dir that is already logged in to Sora (log in once with --verification-prompt, or sign in manually in that profile outside the script). For Cloudflare/login struggles, run once with --verification-prompt and optional --seconds-per-target 30.

Detail-page downloads (default on): after each list page (/profile, /drafts) finishes auto-scroll (and any --seconds-per-target wait), the script scrolls the list again until almost no new links appear, collects every unique detail URL it finds (any element with [href] on sora.chatgpt.com), sorts /p/… posts first, then opens each detail page and resolves URLs only through the site’s Download flow (⋯ → Download, /backend/project_y/download/…, visible Download controls, MP4/download events)—not from <video> preview streams unless you pass --detail-dom-fallback.

No overview/grid streams (default): --passive-media-capture is off by default, so scrolling the profile/drafts grid does not save preview video URLs—only URLs produced after opening each detail and using Download. Use --passive-media-capture if you want the old behavior (record all network media). By default --max-detail-pages 0 = no cap. Set e.g. --max-detail-pages 50 to limit.

On Sora post pages (https://sora.chatgpt.com/p/…), Download is only in the main post ⋮ menu. With comments, the script first restricts ⋮ buttons to those above the “add comment” field (when it can find it), then tries top-to-bottom by position; it scrolls the <video> into view first. Any trigger that doesn’t open Download is closed with Escape before the next. Remaining triggers are tried in a final fallback pass. If you see Download API returned 401, sign in (same --profile-dir) and run capture again.

If Cloudflare appears, run with --verification-prompt, solve it in the browser, then press Enter in the terminal (or fix the session in that profile first).

python sora_downloader.py capture

First-time login (interactive), then reuse profile for unattended runs:

python sora_downloader.py capture --verification-prompt --seconds-per-target 45
python sora_downloader.py capture

If the login UI still does not appear, try Edge or a longer load wait:

python sora_downloader.py capture --browser-channel msedge --goto-wait-until load --login-timeout-ms 120000 --verification-prompt --seconds-per-target 120

Useful options:

python sora_downloader.py capture --auto-max-seconds 20
python sora_downloader.py capture --auto-scroll-steps 80 --auto-step-wait-ms 800
python sora_downloader.py capture --no-auto-explore --seconds-per-target 30

Capture options:

• --targets <url1> <url2> ...
  Pages to visit. Default is profile and drafts.
• --seconds-per-target <int>
  Extra idle seconds after auto-explore per list target. Default: 0 (none).
• --profile-dir <path>
  Persistent Chromium profile directory for login/session reuse. Default: .playwright-profile.
• --login-timeout-ms <int>
  Timeout for opening each target/login page. Default: 60000 (60s). If timeout is reached, capture keeps the page open so you can still complete verification manually.
• --nav-max-attempts <int>
  How many times to retry goto per URL (bootstrap, each list target, each detail) before skipping. Default: 4.
• --nav-retry-delay-ms <int>
  Pause between those retries. Default: 1000.
• --goto-wait-until <commit|domcontentloaded|load|networkidle>
  How long goto waits before continuing. Default: domcontentloaded. Use load if the login UI appears late.
• --browser-channel <chromium|chrome|msedge>
  Engine: bundled Chromium or installed Chrome / Edge. Default: chrome (best for ChatGPT SSO flows).
• --bootstrap-url <url>
  Opened once before targets; default https://sora.chatgpt.com/login.
• --no-bootstrap
  Skip the bootstrap step.
• --headless
  Run without visible browser window.
• --verification-prompt
  Pause for Enter after bootstrap and each target (for login/Cloudflare). Default: off (unattended).
• --no-auto-explore
  Disable automatic scrolling/pagination clicking.
• --auto-scroll-steps <int>
  Max auto-explore scroll steps per page. Default: 40.
• --auto-step-wait-ms <int>
  Delay after each auto step. Default: 600.
• --auto-click-interval <int>
  Try clicking pagination controls every N steps. Default: 4.
• --auto-max-seconds <int>
  Hard time limit for auto-explore per page. Default: 30.
• --no-detail-download-links
  Do not visit detail pages (manifest will only get media if --passive-media-capture is on).
• --passive-media-capture
  Record video URLs from all network traffic (grid + detail). Default: off (detail Download only).
• --detail-dom-fallback
  On detail pages, if Download fails, also scrape <video> / links (may match preview streams like the grid).
• --max-detail-pages <int>
  Max detail pages per list URL; 0 = all links found after gathering (default: 0).
• --detail-gather-max-rounds <int>
  Scroll rounds while discovering links on the list (default: 60).
• --detail-gather-idle-rounds <int>
  Stop gathering after this many rounds with no new links (default: 4).
• --detail-gather-pause-ms <int>
  Delay between gather scroll steps (default: 700).
• --detail-settle-ms <int>
  Wait after loading a detail page before scraping. Default: 2500.
• --detail-download-timeout-ms <int>
  How long to wait for a browser download after clicking Download. Default: 20000.
• Global: --output-dir <path>
  Base folder for manifests/downloads. Default: output.

Outputs:

• output/manifests/capture-<timestamp>.json (captured URLs; entries from detail pages include detail_page when applicable)
• output/manifests/storage-state-<timestamp>.json (auth/session state)

3) Download captured media

python sora_downloader.py download

Optional:

python sora_downloader.py download --manifest output/manifests/capture-YYYYMMDD-HHMMSS.json
python sora_downloader.py download --overwrite
python sora_downloader.py download --upload-dir upload

Download options:

• --manifest <path>
  Use a specific capture manifest. Default: latest manifest in output/manifests.
• --storage-state <path>
  Override auth/session state file used for requests.
• --overwrite
  Replace existing files in the download folder.
• --upload-dir <path>
  Write media files here. Default: <output-dir>/downloads (e.g. output/downloads). Use e.g. upload or C:\Videos\SoraUpload for your upload folder.
• --timeout-ms <int>
  Per-request timeout. Default: 120000.
• Global: --output-dir <path>
  Base folder for manifests and default download location. Default: output.

Press Ctrl+C to stop either command early; the script exits gracefully and keeps partial progress.

Downloaded files go to --upload-dir if set, otherwise output/downloads.

Notes

• If a URL expires, run capture again to refresh signed links.
• Duplicate downloads are automatically skipped using a stable media key (host + path), so refreshed signed URLs with different query params do not redownload.
• Some captures may include .m3u8 playlists. This tool saves them as-is.
• You can customize target pages:

python sora_downloader.py capture --targets https://sora.chatgpt.com/profile https://sora.chatgpt.com/drafts