Original FORMWEFT learning material · 2026-09-14 · CC BY 4.0

Make one useful talking video

You do not need to install a model first. Confirm the message and audio, then check captions and the exported file.

Download templates and checklist (Markdown)

Pick a practical template

Product introduction

One problem, one capability, one next action.

View and copy template

Select the text to copy

Audience: [target customer]
Problem: [specific pain point]
Product: [name]
Verified capabilities: [facts and evidence]
Next action: [one action]
Language / aspect / target duration: [fill in]
Use natural short sentences. Do not invent prices, sales, or outcomes. Flag missing information.

Open this drafting direction

Practical explanation

Teach a method someone can actually follow.

View and copy template

Select the text to copy

Audience: [job role]
Problem: [specific task]
Prerequisites: [materials]
Steps: [no more than three]
Not suitable for: [limits]
Language / aspect / target duration: [fill in]
Explain each action and how to check its result. Avoid jargon and invented benefits.

Open this drafting direction

Customer case

Explain a change using authorized evidence.

View and copy template

Select the text to copy

Before: [authorized, de-identified facts]
Action: [what changed]
Observed result: [evidence, period, measurement definition]
Remaining limitations: [limits]
Next action: [one action]
Language / aspect / target duration: [fill in]
No made-up percentages or endorsements. Do not turn correlation into causation.

Open this drafting direction

Complete the facts before submitting. Opening a template does not submit a request or charge an API. The workbench entry uses a Chinese brief; select your output language there, or copy the English template above.

Start with what you already have

Have a video? Open the workbench, select your local video, and import an SRT file aligned to that recording. Review it, position the captions, and export. Local editing needs neither sign-in nor a video model.
Have an idea? Fill in a template below and use your configured script service to draft it. Read it aloud and shorten awkward sentences. Do not present unverified claims as facts.
Need an avatar? Confirm rights to the likeness and voice first. The optional local H3 workflow may omit words, change speech, or change the character. It is not a word-perfect narration and lip-sync system.

Use the recording as the caption clock

Record or use an authorized narration service after approving the script. Check names, numbers, pronunciation, and pauses before saving the final audio. This guide does not add a narration service to the website.
Obtain caption timings from the actual recording. The workbench imports SRT; audio captions for local H3 also depend on the companion and transcription setup. A script-only timeline is a draft, not verified alignment.
Check the start, middle, and end of each sentence. If the audio changes, align the captions again rather than reusing old timestamps.

Review a small section before producing the full video

Preview a representative 8–15 seconds. Check the voice, captions, character, and aspect ratio before generating the remaining sections.
Use authorized product photos, screenshots, or demonstrations when accuracy matters. Generated appearances, interfaces, and performance claims are not evidence.
When importing multiple clips, verify that the preview contains the assembled duration, not just one clip. Do not publish an incomplete assembly as a full video.

Before you export

Opening and ending: no unintended black frames, clipped speech, or missing segments.
Captions: match the recording; check names, numbers, and currencies. Read them on a phone and keep them away from the mouth and important visuals.
Frame and audio: confirm the aspect ratio and listen to the exported file, not just the editor preview.
Duration: check the actual output length. A target duration is a plan, not a guarantee of narration or model output length.
Export: the workbench records in real time in the browser. Keep the page in the foreground. MP4 is used when supported; otherwise output is WebM. Renaming the extension does not convert the format.
Rights: check footage, likeness, voice, and music permissions. Remove confidential customer information and review your watermark preference.