AI video · checked 2026-10-15Can I vibecode Synthesia?
NOT REALLY· don't bother
The pipeline is a script, a voice, a lip-synced face and a video file, and each part has an open model: Linly-Talker and Duix-Avatar run it locally. What you can't copy is the product around it: licensed actors scanned as avatars, a large voice and language catalogue, studio-grade rendering that doesn't wobble, and SOC 2 reporting that training teams ask for. A local build gets you a stiffer talking head on your own GPU, and several of those open models carry non-commercial or custom licences.
$29/moprice
$348it costs you a year
multi-daybuild time
—people replaced it
What you need on hand.
DifficultyAdvanced · Real infrastructure or OAuth work.
TimeA few days: most of it is getting model weights and the GPU stack to run
- Claude Code, Codex or CursorAny of them can run the prompt.
- An NVIDIA GPUAbout 12 GB of VRAM is comfortable; CPU works but is slow.
- Python 3.11 and FFmpegBoth are installed by the setup script instructions.
- 20 GB or more of diskVoice and lip-sync model weights are large.
Before you buildCheck Synthesia's free Basic plan first: it gives 10 minutes of video a month, but with a watermark and, per its help center, no downloads.
- 1. Install one agent: Claude Code, Codex, Cursor.
- 2. Make an empty folder and open it in the agent.
- 3. Paste the prompt below, press enter, and answer its questions as it builds.
The prompt
Build me a local talking-head video maker to replace Synthesia. Requirements: - Python 3.11 with FastAPI; a single HTML page on http://127.0.0.1:4820 with a script box, a voice picker, a portrait upload and a Generate button. Jobs run one at a time from a queue. - Speech: Piper TTS (piper-tts) with voice files kept in ./models/voices/; a chosen voice is downloaded on first use. Split the script into sentences and write each one to a WAV file. - Face: run Linly-Talker's lip-sync step on the uploaded portrait and the speech WAV. Pin its git commit in setup.sh and read its licence file into the README before first run. - Assembly: ffmpeg joins the clips, burns in subtitles from the script (SRT) and exports ./output/<date>-<title>.mp4 at 1080p, 25 fps. Keep the WAV and SRT next to it. - Storage: jobs, scripts and file paths in SQLite (better-sqlite3 equivalent: sqlite3 module) at ./data/jobs.db; a list page shows status, duration and a preview player. - Limits: refuse scripts over 2,000 characters and portraits under 512 px; run on CUDA when available, otherwise print that CPU rendering takes about 10 minutes per minute of video. - No accounts, no telemetry, local-first; the only downloads are model weights on first run. - Out of scope: multiple presenters per scene, screen recording, translation and a template editor. - README: GPU and driver needs, the model download sizes, the licence caveats and a warning that animating a real person needs their consent.
Opens your agent with the prompt filled in, you press enter. Or copy it and paste it into an empty folder.
Then keep going
Next prompts, same folder.
The first prompt builds the core. Run these one at a time in the same folder to grow it into your own Synthesia.
Add batch mode to the talking-head maker in this folder. Requirements: - Upload a CSV with columns title, script and voice. - Queue one job per row and show progress for the whole batch. - Skip rows whose title already has an output file. - Write a summary CSV of results with durations. - Update the README with the CSV format.
Add caption styling to the talking-head maker in this folder. Requirements: - A settings page for font, size, colour and position, saved in SQLite. - ffmpeg burns subtitles with those settings via an ASS file. - Preview a single frame without rendering the video. - Update the README.
Add language dubbing to the talking-head maker in this folder. Requirements: - Translate the script with the model named in .env (any OpenAI-compatible endpoint). - Pick a Piper voice for the target language and render a second video. - Show both videos side by side on the job page. - Out of scope: lip-sync checking by language. - Update the README.
Add backgrounds to the talking-head maker in this folder. Requirements: - Remove the portrait background with rembg and place the presenter on an uploaded image or colour. - Prepend a 3-second intro image and append an outro. - Keep the original render so a change doesn't re-run the lip-sync. - Update the README.
What you lose
- Licensed professional avatars and their consent paperwork
- 140+ languages and a large voice catalogue
- Templates, screen recording and slide import in one editor
- Brand kit and team workspace, review links
- Studio rendering that holds up at 1080p without visible glitches
Why people still pay
Training and sales teams want a presenter video without a camera crew and are happy to pay for a library of licensed avatars, many languages and a vendor with enterprise security reporting. Editing a line of script and re-rendering is the saved time.
Synthesia or your own copy?
Over
Your usage
24h ×$/h
Synthesia, 1 yr$348
Your own copy, 1 yr$258running + build session
You keep$91pays for itself in 2 months
Estimate · Rendering is free on your own GPU; renting a cloud GPU runs a few dollars an hour. The build is token usage on your agent plan.
Take your data, then cancel.
1 · Take your data out of Synthesia
- 1Open the finished video, click the three dots at the top right and choose Download.
- 2Pick MP4 (1080p), WAV audio, an XLIFF translation file or SRT/VTT captions.
- 3Download before your plan ends: after the billing period the help center says you lose access to your videos.
2 · Cancel Synthesia
- 1Open your profile, then Account info and Billing.
- 2Use the three-dot menu on your plan and choose the cancel option, then confirm.
- 3Self-service works on Starter and Pro; Enterprise contracts go through the account team.
- You keep access until the end of the paid period; a banner shows the date.
- If you cancelled by mistake, contact support within 48 hours; reversal is only 'where possible'.
- Basic (free) accounts can't download videos according to a help-center article we read second-hand through search, so export from a paid plan first.
- The refund policy is on a separate help article we didn't read.
The escape hatch
Don't feel like building it? These are already free.
Linly-TalkerOpen sourceMIT-licensed Python project that turns text or voice into a talking digital human by chaining speech, language and lip-sync models; you supply the GPU.★ 3.5k · active Feb 2026 · self-host, real ops
Linux, WindowsVisit ↗Duix-AvatarOpen sourceOpen toolkit for offline avatar video generation with its own model community licence, so read the terms before commercial use.★ 16k · active Apr 2026 · docker to self-host
Windows, LinuxVisit ↗LivePortraitOpen sourceAnimates one portrait photo from a driving video. It makes the face move and does not write scripts or speech; its licence is custom, so check it.★ 19k · active Jun 2026 · self-host, real ops
macOS, Windows, LinuxVisit ↗
Linux, WindowsVisit ↗Duix-AvatarOpen sourceOpen toolkit for offline avatar video generation with its own model community licence, so read the terms before commercial use.★ 16k · active Apr 2026 · docker to self-host
Windows, LinuxVisit ↗LivePortraitOpen sourceAnimates one portrait photo from a driving video. It makes the face move and does not write scripts or speech; its licence is custom, so check it.★ 19k · active Jun 2026 · self-host, real ops
macOS, Windows, LinuxVisit ↗
The numbers
Synthesia pricing
| Plan | Monthly | Annual, per month | What you get |
|---|---|---|---|
| Basic | $0 | $0 | 500 credits a month, 10 minutes of video or dubbing, 1 editor, watermark on videos. |
| Starter | $29 | $18 | 1,250 credits a month (monthly) or 15,000 a year (annual); about 12 minutes of video a month; 1 editor and 3 guests. |
| Pro | $89 | $64 | 6,000 credits a month (monthly) or 72,000 a year; about 60 minutes of video a month; 1 editor and 5 guests. |
| Enterprise | custom | — | Custom credit volumes, unlimited video minutes, custom seats. |
- Free tier
- 10 minutes of video a month, watermark, no downloads.
- Billing
- Monthly plans can be cancelled any time. Annual billing is $18 a month for Starter and $64 for Pro, charged yearly. The price shown here is the monthly-billed rate.
- Hidden costs
- Video costs 100 credits a minute and dubbing 76, so the monthly credits cap output; the page says the first 10 minutes on any plan are included. Our source for rollover was a third party (unused credits do not roll over), unverified. Custom studio avatars cost extra.
- Source
- synthesia.io
Also on the list
Related apps
Basic is free: 10 minutes of video or dubbing a month, 3 survey responses, 500 credits, and videos carry a Synthesia watermark.
Not really. Training and sales teams want a presenter video without a camera crew and are happy to pay for a library of licensed avatars, many languages and a vendor with enterprise security reporting. Editing a line of script and re-rendering is the saved time. The prompt on this page builds the closest honest substitute.
About $29 a month on the Starter plan, $348 a year (checked 2026-10-15).
Licensed professional avatars and their consent paperwork; 140+ languages and a large voice catalogue; Templates, screen recording and slide import in one editor; Brand kit and team workspace, review links; Studio rendering that holds up at 1080p without visible glitches. If any of those matter to you, keep paying.
Yes: Linly-Talker (open source), Duix-Avatar (open source), LivePortrait (open source).