Send YouTube Transcripts to Hugging Face
Extract any YouTube transcript free, then feed it into Hugging Face
Or just change youtube.com to 2outube.com in your browser
Swap 'youtube.com' for '2outube.com' in any video URL to get the full transcript instantly—no signup required. Copy the text and send it directly to Hugging Face models, datasets, or Inference API pipelines for summarization, classification, or fine-tuning.
The Trick
youtube.com/watch?v=VIDEO_ID
2outube.com/watch?v=VIDEO_ID
Just change 'y' to '2'
Works with any YouTube video that has captions
Using Transcripts with Hugging Face
Get the YouTube transcript
Navigate to any YouTube video you want to analyze. Replace 'youtube.com' with '2outube.com' in the URL and press Enter. The full transcript appears immediately—no login, no extension, no copy-pasting from captions.
Copy the transcript text
Select all transcript text from the 2outube page. You can copy the plain text directly or use the copy button if available. The output is clean, timestamped or plain depending on your needs.
Send to a Hugging Face model or dataset
Paste the transcript into the Hugging Face Inference API playground, a Spaces app, or programmatically via the `transformers` pipeline or `datasets` library. Use it as input text for summarization (e.g. facebook/bart-large-cnn), zero-shot classification, embeddings, or push it as a dataset row for fine-tuning.
Run your HF pipeline
Whether you're building a transcript summarizer, training a custom classifier on lecture content, or extracting named entities from video interviews, the raw transcript from 2outube drops straight into any HF pipeline with no preprocessing overhead.
Quick Start
Get the transcript
Open the YouTube video you want. Copy the URL from your browser address bar.
Change youtube to 2outube
In the URL, replace 'youtube.com' with '2outube.com'—literally just change the 'y' to a '2'. Hit Enter and the transcript loads instantly.
Paste into Hugging Face
Copy the transcript and paste it into the Hugging Face Inference API widget, a Gradio Space, or your Python script using the `transformers` or `datasets` library for summarization, classification, embedding, or fine-tuning.
Ready-Made Template
# Step 1: Get transcript from 2outube (manual copy or automate with requests)
# URL pattern: https://2outube.com/watch?v=VIDEO_ID
from transformers import pipeline
# Paste your transcript here
transcript = """
[Paste transcript from 2outube here]
"""
# Summarize with BART
summarizer = pipeline("summarization", model="facebook/bart-large-cnn")
# Chunk if transcript is long (BART max ~1024 tokens)
chunk_size = 1000
words = transcript.split()
chunks = [" ".join(words[i:i+chunk_size]) for i in range(0, len(words), chunk_size)]
summaries = [summarizer(chunk, max_length=150, min_length=40, do_sample=False)[0]["summary_text"] for chunk in chunks]
final_summary = " ".join(summaries)
print(final_summary)
# --- Push to HF Dataset ---
# from datasets import Dataset
# ds = Dataset.from_dict({"video_id": ["VIDEO_ID"], "transcript": [transcript], "summary": [final_summary]})
# ds.push_to_hub("your-username/youtube-transcripts")Questions
Does this work with any YouTube video?
Yes, any video with captions—auto-generated or manually added. If a video has no captions at all, no transcript is available anywhere.
Is it really free?
Completely free. No account, no limits, no API key required for 2outube.
What Hugging Face tasks work best with YouTube transcripts?
Summarization (BART, Pegasus, T5), zero-shot classification, named entity recognition, sentiment analysis, text embeddings for semantic search, and question-answering all work well. Long transcripts may need chunking before passing to models with token limits.
How do I handle long transcripts that exceed HF model token limits?
Split the transcript into chunks of ~800–1000 words before passing to the model. For summarization, summarize each chunk and then summarize the summaries. The `transformers` library's `pipeline` handles this with the `truncation=True` flag for quick tasks.
Can I build a Hugging Face dataset from YouTube transcripts?
Yes. Collect transcripts via 2outube, structure them as a list of dicts with fields like video_id, title, transcript, and label, then use `datasets.Dataset.from_dict()` and `.push_to_hub()` to publish your dataset to the Hugging Face Hub.
Can I automate fetching transcripts to feed into a HF pipeline?
2outube is a browser-based tool. For automated pipelines, you can use the `youtube-transcript-api` Python library to fetch transcripts programmatically, then pass them directly into your Hugging Face pipeline. 2outube is ideal for quick one-off extractions.
Does 2outube preserve timestamps in the transcript?
Yes, timestamps are included in the transcript output. You can strip them for clean text input to HF models, or keep them if you need time-aligned data for tasks like audio-text alignment or temporal analysis.
What languages are supported?
Any language that YouTube provides captions for. If the video has auto-generated captions in Spanish, French, German, Japanese, or another language, 2outube returns that transcript. Hugging Face has multilingual models like mBART and XLM-R that can process non-English transcripts.
Extract a YouTube Transcript for Hugging Face — Free
Free, no signup required
Try It Free