DEV Community: fguiraud · Data Tools

Transcribe TikTok and Instagram videos to text with Python (Whisper, no API key)

fguiraud · Data Tools — Thu, 01 Oct 2026 03:54:38 +0000

Short videos are where a lot of today's marketing, product feedback and trends live, but the words are locked inside the video. TikTok and Instagram don't give you the transcript, and most videos have no captions.

The usual do-it-yourself route has three parts: download the video (TikTok and Instagram block many server IPs), extract the audio, and run a speech-to-text model. In this post I'll do all three with one API call from Python, and then go one step further: transcribe a whole TikTok account and rank its videos by views, with the first sentence of each one (the "hook").

Disclosure: I built the tool used here, Video to Text Transcriber on Apify. It runs open-source Whisper and costs $0.006 per minute of video; videos with only music are not charged. Apify's free plan includes monthly credits.

Setup

pip install apify-client
export APIFY_TOKEN=your_token

Transcribe TikTok and Instagram links

import os
from apify_client import ApifyClient

client = ApifyClient(os.environ["APIFY_TOKEN"])
run = client.actor("fguiraud/video-to-text-transcriber").call(run_input={
    "sources": [
        {"url": "https://www.tiktok.com/@tiktok/video/7689568478570843422"},
        {"url": "https://www.instagram.com/reel/C0hQSaMpD97/"},
    ],
    "outputs": ["text"],
})

for item in client.dataset(run.default_dataset_id).iterate_items():
    v = item.get("video") or {}
    views = f"{v['views']:,} views" if v.get("views") else "views not public"
    print(f"{v.get('platform')} | {v.get('author')} | {views} | {item['billedMinutes']} min")
    print("  ", item.get("text", "")[:110], "...")

Real output (October 2026):

tiktok | tiktok | 335,300 views | 1 min
   You three look so good. I'm getting serious 2000 flashbacks. Thanks. I have this disposable camera. If I were  ...
instagram | NASA Goddard | views not public | 1 min
   Hey Space Nerds, I got a special delivery for you. So picture this. You're on a military range in the middle o ...

Instagram doesn't make view counts public for every Reel, so that field can be empty; likes and comments are there. X (Twitter) and Facebook video links work the same way.

A whole TikTok account, ranked by views

Instead of collecting links by hand, give the account. This saves the newest videos of @nasa to a CSV with their date, views, likes and the first sentence of what is said:

import csv
import os
from apify_client import ApifyClient

client = ApifyClient(os.environ["APIFY_TOKEN"])
run = client.actor("fguiraud/video-to-text-transcriber").call(run_input={
    "profiles": ["@nasa"],          # or https://www.tiktok.com/@nasa
    "maxVideosPerProfile": 5,
    "onlyNewVideos": False,         # True on a weekly schedule: only new videos are transcribed and billed
    "outputs": ["text"],
})

rows = []
for item in client.dataset(run.default_dataset_id).iterate_items():
    if item.get("status") != "ok" or not item.get("text"):
        continue  # music-only videos come back as "no speech" and are not billed
    v = item["video"]
    hook = item["text"].split(". ")[0][:90]   # the first sentence: what grabs attention
    rows.append({"date": v["uploadDate"], "views": v["views"], "likes": v["likes"], "hook": hook, "url": v["url"]})

rows.sort(key=lambda r: r["views"] or 0, reverse=True)
with open("nasa_tiktok.csv", "w", newline="", encoding="utf-8") as f:
    writer = csv.DictWriter(f, fieldnames=rows[0].keys())
    writer.writeheader()
    writer.writerows(rows)
for r in rows:
    print(f"{r['date']}  {r['views']:>9,} views  {r['hook']}")

Real output (October 2026):

2026-09-18    243,300 views  From the skies over Pittsburgh to the path back to the moon, NASA's mission kept moving fo
2026-09-17    163,800 views  NASA's reusable shuttle orbiter, part of a new, less expensive space transportation system
2026-09-25    105,900 views  This week NASA inspired future explorers, prepared the next crew for launch, studied life

Five videos were requested and three rows came back: the other two had only music, so they were marked as having no speech and were not charged.

Track an account every week

Set "onlyNewVideos": True and run the script on a schedule (cron, or an Apify schedule on the Actor itself). Videos transcribed in earlier runs are skipped, so each week you pay only for what was posted since. That gives you a running log of what a creator or a competitor says, ready for a spreadsheet or an LLM summary.

Subtitles instead of text

Ask for "outputs": ["srt", "vtt"] and you get subtitle files with readable line lengths (42 characters, 2 lines), ready to re-upload with your own videos.

Cost

$0.006 per started minute of video, so a typical 15 to 60-second TikTok or Reel costs $0.006 and an account with 30 short videos about $0.18. Music-only videos and links that can't be downloaded (private videos, for example) are not charged. Instagram profiles can't be listed without a login, so for Instagram pass the Reel links.

Tips

Paste mixed lists: TikTok, Instagram, X and Facebook links, and direct MP4/MP3 files, all in one run.
Set "language": "es" (or any code) when you know the language; otherwise it's detected automatically.
Use "model": "small" for noisy audio or accents ($0.012 per minute, more accurate).

Questions or ideas for what to do with this data? Leave a comment.

YouTube transcripts in Python for LLMs, without getting blocked

fguiraud · Data Tools — Mon, 28 Sep 2026 11:00:00 +0000

Getting the transcript of a YouTube video is easy on your laptop. Run the same code on a server, a notebook in the cloud or a scheduled job, and after a few requests YouTube answers with:

Sign in to confirm you're not a bot

That's YouTube blocking data-center IPs. I hit it myself after a handful of requests from a cloud machine while building this. The usual fix is rotating residential proxies, which is a small project on its own.

In this post I'll get transcripts from Python with that part handled on the server side, in formats that are ready for LLMs.

Disclosure: I built the tool used here, a YouTube Transcript Scraper on Apify. It costs $0.003 per transcript (residential proxies included); videos without captions are not charged. Apify's free plan includes monthly credits.

Setup

pip install apify-client
export APIFY_TOKEN=your_token

Transcripts as Markdown with timestamps

import os
from apify_client import ApifyClient

client = ApifyClient(os.environ["APIFY_TOKEN"])
run = client.actor("fguiraud/youtube-transcript-scraper").call(run_input={
    "videos": [
        "https://youtu.be/arj7oStGLkU",
        "https://www.youtube.com/watch?v=dQw4w9WgXcQ",
        "https://youtube.com/shorts/U5-dE2nCzjM",
    ],
    "languages": ["en"],
    "outputs": ["markdown", "segments"],
})

for v in client.dataset(run.default_dataset_id).iterate_items():
    if v["status"] != "ok":
        print(f"[{v['status']}] {v['url']}")
        continue
    kind = "auto" if v["isAutoGenerated"] else "human"
    print(f"{v['title']} ({v['language']}, {kind} captions, {v['wordCount']} words)")

Real output (September 2026):

[no-captions] https://www.youtube.com/watch?v=U5-dE2nCzjM
Rick Astley - Never Gonna Give You Up (Official Video) (4K Remaster) (en, human captions, 487 words)
Inside the Mind of a Master Procrastinator | Tim Urban | TED (en, human captions, 2277 words)

The Short has no captions (common for Shorts and music without speech), so it's reported and not charged.

The Markdown puts a timestamp at the start of each paragraph:

# Inside the Mind of a Master Procrastinator | Tim Urban | TED

**[00:00:12]** So in college, I was a government major, which means I had to write a lot of papers. ...

That matters for LLMs: when you ask for a summary, the model can cite where in the video each point is.

Summarize with an LLM

prompt = f"Summarize this talk in 5 bullet points, citing timestamps:\n\n{v['markdown']}"
# send `prompt` to the model of your choice

For long videos and RAG, ask for "chunks": ~1,000-character pieces with start and end times and a token estimate, ready to embed.

Other formats and languages

"outputs": ["srt"] or ["vtt"] gives subtitle files for video editors.
"languages": ["es", "en"] returns Spanish captions when they exist and English otherwise (es also matches regional tracks like es-419).
Human captions are preferred over auto-generated ones; availableLanguages lists every track of the video.

What it doesn't do

It reads the captions YouTube already has. It doesn't translate into languages the video has no captions for, and it can't read private, members-only or age-restricted videos (those come back as unavailable, not charged). For your own audio or video files without captions, a Whisper-based transcriber is the right tool.

Scripts (Markdown files, SRT, bulk lists from a .txt) are in the data-tools repository. Questions or feature ideas? Comments are open.

How to get Google Trends data in Python without 429 errors

fguiraud · Data Tools — Sat, 26 Sep 2026 19:32:36 +0000

Google Trends is one of the best free signals for market research, SEO and content planning: is interest in a product growing, in which country, and what are people starting to search for? The problem is getting that data into Python reliably.

There is no official API. Unofficial libraries call the same endpoints the website uses, and Google answers bursts of requests with HTTP 429 (Too Many Requests). On a laptop you retry and wait; in a scheduled job or a data pipeline, a 429 means a failed run.

In this post I'll show a way to get the data with a few lines of Python, where the retries and IP rotation happen on the server side.

Disclosure: I built the tool used here, a Google Trends Scraper on Apify. It is pay-per-use ($0.002 per search term); Apify's free plan includes monthly credits, enough for everything in this post.

Setup

pip install apify-client
export APIFY_TOKEN=your_token   # Apify Console → Settings → API & Integrations

Compare terms and see which one is growing

import os
from apify_client import ApifyClient

client = ApifyClient(os.environ["APIFY_TOKEN"])
run = client.actor("fguiraud/google-trends-scraper").call(run_input={
    "searchTerms": ["chatgpt, gemini, claude"],  # one line = one comparison, same 0-100 scale
    "geo": ["US"],
    "timeframe": "today 12-m",
    "outputs": ["interestOverTime", "relatedQueries"],
})

for item in client.dataset(run.default_dataset_id).iterate_items():
    for term, s in item["summary"].items():
        print(f"{term:10} {s['trend']:8} change {s['changePercent']:+.0f}%  peak {s['peakDate'][:10]}")

Real output (September 2026):

chatgpt    falling  change -21%  peak 2025-10-19
gemini     rising   change +28%  peak 2026-04-19
claude     rising   change +228%  peak 2026-05-24

The summary answers the question you usually have ("is it growing?") without writing the analysis yourself: it compares the last fifth of the period with the first fifth, which is robust to single noisy weeks, and leaves out the current incomplete week so it never looks like a drop.

The full weekly series is in item["interestOverTime"], one point per date with a value per term, ready for pandas:

import pandas as pd

df = pd.DataFrame([{"date": p["date"][:10], **p["values"]} for p in item["interestOverTime"]])
df.set_index("date").plot(title="Search interest, US")

Rising searches: what people are starting to look for

This is the most useful part for SEO and content: related searches that are growing fast, before keyword tools catch up.

for term, related in item["relatedQueries"].items():
    for q in related["rising"][:3]:
        print(f"[{term}] {q['query']} ({q.get('label')})")

[chatgpt] chatgpt caricature (+3,300%)
[gemini] gemini spark (+400%)
[claude] claude cowork (Breakout)

"Breakout" means more than +5,000%.

What's trending today

run = client.actor("fguiraud/google-trends-scraper").call(run_input={"trendingNow": ["US", "GB"]})
for row in client.dataset(run.default_dataset_id).iterate_items():
    print(row["geo"], row["rank"], row["query"], row["approxTraffic"])

Each trending search comes with its approximate traffic and the news articles behind it.

Tips

Values are relative. 100 is the peak for the terms compared together in that period, so only compare numbers from the same query.
Rare terms return no data in small countries or short periods. Those come back as status: "no-data" and are not charged.
Weekly alerts: schedule the Actor in Apify and turn on Monitor new rising searches to receive only the rising searches that appeared since last week.
You can also paste a trends.google.com link you already use as a search term: it keeps its countries, dates and filters.

Wrapping up

If you only need a few manual lookups, the Google Trends website is fine. When Trends data feeds a report, a dashboard or an AI agent, you want something that doesn't fail on the third request. Ready-to-run scripts (CSV export, trending now) are in the data-tools repository.

Questions or ideas for the tool? Leave a comment, I read all of them.