# Make one useful talking video

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

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

## 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 Do not publish an incomplete assembly as a full video.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.

## Pick a practical template

### Product introduction

One problem, one capability, one next action.

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.

### Practical explanation

Teach a method someone can actually follow.

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.

### Customer case

Explain a change using authorized evidence.

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.

Use: these original FORMWEFT templates and checklists may be reused with attribution under CC BY 4.0. They are documents, not executable software, models, voices, or third-party assets. They do not certify or install third-party tools.

## Further reading and inspiration

Inspired by audio-first sequencing and optional avatars. No source article text, packages, or character assets were copied. The tools and commercial terms discussed in those articles have not all been verified.

- [X榜单 · 文章索引](https://xbangdan.com/articles/)
- [黄小木 · 本地数字人口播工作流](https://x.com/ai_xiaomu/status/2089324371991216173)
- [Miles Ma · 文章到视频的制作流程](https://x.com/miles_mazy/status/2097177704282136838)