Synthesia Text to Video: Scripts, Prompts and PowerPoint (2026)

Synthesia turns text into a presenter-led video: your script is split into scenes and an AI avatar reads it, with voices in 140+ languages and accents. There are six ways in, and the route decides whether your exact words survive. Import script keeps your text word for word, while Import a file writes a new script from the document. Prompts, script imports and file imports run through Assistant, which needs a paid plan, so Basic users start from a template, a blank project, a PowerPoint deck or a screen recording. Starter is $19 a month billed monthly, or $14 a month billed yearly ($168 a year) ^[1]. Each scene holds up to 5 minutes of script. Plan around that.

Six starting points, and which ones keep your words

The split that matters is between routes that keep your wording and routes that rewrite it. A compliance script approved by legal belongs in Import script or a blank project, never in Import a file, because the help docs say a file import produces a new script and does not keep the original text. Prompt and URL routes are for first drafts. PowerPoint sits in between: notes are read as written, while slides without notes get a script generated from the slide text. Decide this first.

Pasting a finished script without losing a word

Import script is the route for text you have already approved. Assistant uses it without rewriting, and you start a new scene by leaving two soft returns (Shift+Enter twice) between sections; the legacy assistant needed only one ^[2]. Scene breaks are not cosmetic. Scripts are scene-based, a single scene cannot exceed 5 minutes of script, and longer text must be split ^[3]. With up to 150 scenes, one video can run up to 4 hours, so length is rarely the constraint; scene structure is.

Inside a scene, a few controls shape the delivery. Pauses default to 1 second and have no count limit, so you can slow a dense passage without rewriting it. Synthesia detects the script language automatically, and one avatar can speak several languages in one scene, but the voice changes only between lines, not mid-sentence. Voice speed runs from about 0.8x to 1.2x, per paragraph or for the whole video. Rebuilt pronunciation controls, on all plans, let you set a word by speaking it and apply the fix across the whole video in one click. On Enterprise, saving it to the workspace Glossary applies it across all voices in that language. Fix names once.

Letting Assistant draft from a prompt, URL or document

Assistant, the chat-based tool launched in July 2026, is the route when you have material but no script. It is listed on Starter, Pro and Enterprise. It accepts a prompt, URLs and one or several files up to 50 MB each in ppt, pptx, pdf, doc, docx or txt format, or a mix of all three. Before generating anything, it shows a numbered outline of sections you can edit, delete or add to; then you either generate the video directly or continue to the editor.

Two settings shape the draft. Duration is Auto, Short, Medium or Long; Synthesia says Auto was tested for 1-5 minute videos and falls back to Medium when the prompt names no length. Delivery style is Dynamic, with shorter scenes and frequent cuts, or Presentation, with longer, denser scenes. There is no language setting. The script follows the language of your prompt, or of the attached files when there is no prompt, unless the prompt names a language. Write the prompt in the language you want spoken.

In Storyboard, each scene then offers quick actions: Improve writing, Shorten, Longer and Change tone. Assistant itself uses no credits; generating the video does. Synthesia keeps prompts for abuse monitoring for up to 30 days, and the legacy AI Video Assistant, in service since January 2024, is still reachable until Assistant replaces it. Our guide to Assistant and Syren compares the two in depth.

Turning a PowerPoint deck into scenes

PowerPoint import is on every plan, Basic included, and it does not go through Assistant. It takes .pptx files up to 1 GB or 150 slides ^[4]. Each slide becomes a scene with an avatar narrator, speaker notes become the voiceover, and slides without notes get a script built from the slide text. Text, shapes, images, videos and tables become editable elements; tables turn into motion graphics, and 4:3 decks are converted to 16:9 automatically.

Some things do not survive the trip. PowerPoint animations are not imported, although an Add animations with AI option places AI-selected animations on key elements. Basic and Starter users cannot upload custom fonts for imported decks, so a brand typeface falls back to the fonts the editor offers. The pricing page lists custom font uploads on Pro and Enterprise; Synthesia's help center describes them as Enterprise-only, so confirm before buying Pro for fonts. Write the speaker notes as narration before you upload. They are read as written, which turns the deck itself into your script review before anyone opens Synthesia.

Since October 2026, importing from the Home page offers a choice. Keep my slides imports each slide as an editable scene, as before. Redesign with Syren hands the deck to Syren, which rebuilds it as a dynamic video instead of Studio scenes. That choice is hard to undo. Syren videos live apart from Studio videos and cannot be opened in the scene editor, so a redesigned deck cannot later be fixed slide by slide. Updating quarterly? Keep the slides.

Screen recordings and Syren, when your text is not a script yet

Software walkthroughs usually start from a recording rather than a document. External MP4 uploads up to 500 MB and 30 minutes are transcribed into an editable script split into scenes, on all plans, and the AI Screen Recorder extension for Chrome and Edge is available on Basic too. From there the transcript behaves like any other script: tidy the wording, add pauses, and let an avatar voice read it over the footage.

Syren, the AI video agent launched in October 2026 and labeled beta, works the other way round. You describe the video, attach decks, PDFs, images, video, audio or YouTube links, and Syren builds motion graphics, avatars, voiceover, music and b-roll, which you refine by chatting. Videos run up to 60 minutes on paid plans and 3 minutes on Basic, where the help center says Syren videos are watermarked and can be downloaded and shared only for the first 2 weeks. The pricing page says a 1-minute Syren video typically uses 1,000-2,000 credits, against about 100 credits per finished minute for a Studio avatar video on self-serve plans. Synthesia itself positions Syren for fast creative work and the editor for exact scripts.

Since November 2026, an MCP connection in public beta lets you start from ChatGPT or Claude, creating a Studio draft from a prompt or from an existing script with the narration kept verbatim. Connecting works on every plan, but creating a Studio draft needs a paid plan, and Synthesia's docs warn that these drafts may be rougher than Assistant's. Hourly limits apply.

Generating: credits, moderation and the Basic ceiling

Nothing is charged while you write, edit or preview. Credits are spent on the avatar footage generated, and only new footage counts, so re-rendering one corrected sentence costs that sentence, not the whole video. Every plan gets 10 minutes of video or dubbing, 3 survey responses and 500 credits a month; by our arithmetic, Starter's 1,250 monthly credits add up to about 12 more minutes and Pro's 6,000 about 60 more, at roughly 100 credits per finished minute. Pro is $89 a month billed monthly, or $59 a month billed yearly ($708 a year). Allowances reset every billing period with no rollover; our pricing breakdown covers top-ups and billing rules.

Every script is moderated at generation by automated systems plus human review. Stock avatars are limited to neutral, factual content, so scripts touching medical, legal, financial, news or political topics, or naming real people or companies, can be ruled ineligible for a stock avatar. On Basic, Starter and Pro, a script-remediation checker flags issues before you generate, but its guidance is advisory. Paid plans get one appeal per rejection. Basic gets none.

Want your own face reading the script? Personal Avatars are included on Starter (3) and Pro (5). Synthesia creates a personal avatar or voice clone only with a live consent recording from that same person, who must be of legal age; it refuses celebrities, public figures, deceased people and anyone who has not consented, and you cannot create a photo avatar of someone else even with their permission. Stock avatars cannot appear in paid advertising without Synthesia's written consent.

Basic videos carry a Synthesia watermark and are shared by link or embed, with no MP4 download. MP4 download at 1080p starts on Starter, and videos with interactive buttons or branching never download as MP4 on any plan, because they play only in the Synthesia player. If the goal is a file for an LMS upload or a town-hall slide, budget for Starter from day one.

The bottom line

Synthesia's text-to-video works best when the text already exists and must stay exact: paste it through Import script or a blank project, break it into scenes under 5 minutes, and fix product names once with the pronunciation controls (or, on Enterprise, in the workspace Glossary). Use Assistant when you only have source material, PowerPoint when the deck already is the course, and Syren when speed matters more than control. Basic is enough to test templates, decks and recordings. Anything with a prompt, a file import or an MP4 at the end starts at Starter.