Script2Film does not synthesize voices, generate video or edit by itself. It splits a film into stages, connects an existing tool to each, and passes work between stages only as files in a folder. That is why any stage can be swapped for the tool you prefer.
The first example uses only this part. You install the models and packages yourself; the install check tells you what is missing.
Local speech synthesis. About ten seconds of reference audio is enough to match your voice, or you can design a synthetic voice from a text description.
Transcribes the finished narration back to text and measures when every word is spoken. Take picking, subtitles and the timing of visuals all depend on it.
Each of the 27 scenes is a web animation that HyperFrames captures frame by frame into video. The scene code is in this project.
Joins narration, visuals and subtitles into one film. Every exported film is measured for length, black frames, loudness and whether the sound effects made it into the mix; explainers also get the check sheet.
Background music is written by the project's composer program from the on-screen events and rendered offline with a SoundFont. Sound effects are Kenney's CC0 assets and some made in-house.
The fonts used for visuals and subtitles. The license text ships next to the font files. Two optional fonts, LXGW WenKai and jf open-huninn, are listed with their original download links.
All of them are optional: you can make a film without them. When you connect one, the tool works inside the account you logged in to yourself: submit, wait, collect, check. The agent asks before spending credits or compute units, and you type your own account name and password.
Video from text or images. Script2Film works out how long each shot must be, writes the prompts, lays them out as a run sheet, submits and collects them, then measures the clips and checks for bad first and last frames. The project also includes a full set of prompt-writing rules and storyboard tools for VideoExpress.
One portrait plus text or audio makes a lip-synced host. Script2Film cuts the narration into segments, uploads the look image, submits each segment, collects the results and places them back on the timeline as a circle, a side frame or full frame.
An open-weight video model that generates on the Colab GPU of your own Google account. The narration recording drives the host, so the lips follow it, and the prompt can make the person gesture, walk while talking, or film a handheld selfie vlog. It can also generate shots and extend them. Script2Film cuts the narration into parts of 15 s or less, writes the host prompts, checks which country the machine is in before booting, and puts the results back on the timeline.
An open-weight music model. Give it lyrics and a description of the style, and it makes a song with vocals or an instrumental; it can sing in Taiwanese Hokkien and Mandarin. Script2Film writes the job, boots the GPU, generates, downloads and releases the machine, then sets the WAV as the film's background music; you can also paste the code into a Colab notebook yourself.
Disclosure: the VideoExpress and TalkingPhotos buttons, and the CloneVoice.ai link in the narration card above, are referral links. The author uses VideoExpress and TalkingPhotos every week. If you buy through a link the author receives a small commission, and the price you pay is the same (prices as of November 2026). The core flow works fully without either.
If you want to get better at writing for video, try writing by hand with this small tool that comes with the project.
Helps you expand an idea into a storyboard and prompts, for practice or to compare with a storyboard an agent wrote. Its patterns are tuned for VideoExpress. How to split shots, describe characters and write dialogue is worth a look with other AI video tools, but none of it has been tested on any other platform.
A free, open-source video editor. Script2Film writes an edit plan and sends it straight into Concat: the host frame, keywords, chapter labels and music are all on the timeline, and you can keep adjusting.
Upload, scheduling, thumbnail, description and the altered-content label. You complete the authorization yourself in the browser, and whether to upload and when to publish is entirely up to you.
What each stage reads and writes, and how to name files when you use another platform, is in the flow document. The license of every package and asset is listed in the notice file.