When you create an SOP from a video, the AI uses both the picture and the sound as clues to create subtitles, step splits, and step body text. The more clues you give the AI, the higher the accuracy.
This article explains what the AI looks at and creates by default, and the four places where you can add clues (Terminology Glossary, Authoring rules, Analysis instructions, and how you speak while filming), in order from the widest setting scope to the narrowest.
What the AI does by default
Even when you analyze a video without changing any settings, the AI reads both the picture and the sound of the video and creates the following.
| What it creates | Created from |
|---|---|
| Action subtitles (AI-generated) | Created from the picture. The movements of hands, tools, and parts are turned into subtitles timed to when the movement appears on screen. In scenes where someone is only talking and no hands are moving, the main point of the talk is made into a subtitle in a form such as "Verify ~". |
| Speech subtitles (Speech transcription) | Created from the sound. What was said is turned into subtitles, written in written-language form while keeping the sentence structure. Hesitations such as "um" are removed, and model numbers and proper nouns are kept as they are. Parts that cannot be heard clearly are not guessed and are not made into subtitles. |
| Step split | Decided from both the picture and the sound. The AI judges by combining breaks in the picture, such as the tool or work target changing or a hand leaving a part, with changes in the talk, such as "Next, ~" or "This completes ~." (When Split granularity is set to "Action", the steps are split for each action.) |
| Step body text | Written from both the picture and the sound. The actions in that section are combined with the instructions, cautions, and numbers that were spoken, and placed in the body text, work tips, and notes. |
The two types of subtitles are created separately, and you choose which one to use as subtitles in "Step conversion settings" when you create the SOP. By default, only "AI-generated" (action subtitles) is ON. If you want the speech spoken during filming to appear as subtitles, also check "Speech transcription" (see "5. Choose which subtitles to use when creating an SOP" below).
Transcription takes the context of the whole video into account. However, internal names used only within your company, how model numbers are written, and numbers or judgment criteria that do not appear on screen cannot be read correctly from the picture and sound alone. The four places below are there to fill this gap.
Four places to add clues
| Location | Setting scope | Mainly affects | Plan |
|---|---|---|---|
| 1. Terminology Glossary | Once for the whole company | Term spelling, step splits | Term registration: Core, Pro, Enterprise |
| 2. Authoring rules (How to write it) | Once for the whole company or each site | Wording of action subtitles and body text | Enterprise |
| 3. Analysis instructions | Per video | Understanding of what work is being done (subtitles, splits, body text) | All plans |
| 4. How you speak while filming | At the time of filming | Speech subtitles, step splits, body text | All plans |
They are listed from top to bottom, starting with the settings that apply to all later analyses once you set them up one time. You do not need to set up all of them; even one alone will help.
1. Register internal terms and site assumptions in the Terminology Glossary
Set this in the "Dictionary / Templates" → "Terminology Glossary" tab in the sidebar. If you register content shared across the company once, it is automatically applied to all later AI analyses, so you do not need to enter it for each video.
Terms
Register the names used inside your company and the standard spelling for product names, equipment names, part names, process names, and so on. The AI gives priority to the registered terms and writes them in that spelling in subtitles and body text. It is especially effective for terms that are easily mistaken for common words with similar sounds, and for terms such as model numbers whose spelling cannot be determined from sound alone.
If you add a "Description (reading / meaning)" to a term, recognition accuracy improves further. The description is read by the AI during analysis and is not used for translation.
You can also load terms from industry templates or register them in bulk from Excel / CSV. For details, see Terminology Glossary (terms and multilingual translations). Term registration is available on the Core, Pro, and Enterprise plans.
About this site
In the same tab, write the industry, business, product overview, and main process types under "About this site." The AI uses this as the premise when analyzing videos, which improves the accuracy of how terms are interpreted and where steps are split. Only one is set for the whole company (editing is available to Team Administrators and above).
2. Align wording with "How to write it" in Authoring rules
If you write your organization's wording and Prohibitions in "How to write it" in the "Dictionary / Templates" → "Authoring rules" tab, the text the AI creates will follow them.
- Action subtitles (AI-generated): written according to the wording and Prohibitions in "How to write it"
- Step body text: If you select "How to write it" for "Text style" in the AI processing options, the text is adjusted to the wording and granularity defined in "How to write it"
"How to write it" is not for reading the spoken content or actions correctly. It is for aligning how the read content is written. Use it when the content is correct but the wording differs from your company's writing style. Note that "Regulations" in Authoring rules is used for checking in the Pre-publish check, and "How to write it" is what takes effect when writing text.
Authoring rules is a feature of the Enterprise plan. For details, see Setting SOP writing rules (reflecting your organization's wording and standards in the AI) and Summary of what to follow when creating an SOP (Authoring rules).
3. Describe the work in the analysis instructions
In "Analysis instructions" in the AI processing options that open when you upload, write what work the video shows. Writing the target equipment, the purpose of the work, the tools used, and so on gives the AI clues for reading the picture. In addition to the type of work, you can also instruct it on scenes to keep, how to write, and how questions are created (Advanced use of analysis instructions: keep NG and redo scenes to create training materials (customer service role-play example)).
You only need to write what is specific to that video. Leave terms and assumptions shared across the company to the Terminology Glossary. For example, you can write it like this.
Target equipment: XXX / Main purpose: Replacing XXX / Tool used: Torque wrench / Caution: Do not reuse XXX
From the icon at the top right of the field, you can insert a registered template. If you align the format of each item, the amount of explanation will not vary depending on the person writing it. The purpose, readers, and Scope decided in AI consultation before you build (Core, Pro, Enterprise plans) are automatically carried over to this field.
4. Speak briefly in step with the work while filming
What you say while filming is used for speech subtitles, step splits, and step body text. Speaking short sentences that match the work is more effective than saying single words. Single words do not connect what you are doing or how, so they are hard to use as clues for the body text or the splits. You do not need long explanations.
| What you say | Example | Mainly affects |
|---|---|---|
| What you are about to do | "Next, remove the cover." | Step split (used as a switch point in the talk) |
| Names of parts and tools | "Insert XXX into YYY." | Spelling in subtitles and body text |
| Numbers and pass/fail criteria | "If the gap is within 0.1 mm, there is no problem." | Body text, work tips and cautions |
| Caution and its reason | "Tighten only until it stops by hand, because overtightening will crack it." | Body text, work tips and cautions |
| End of a section | "This completes the installation of XXX." | Step split |
Things to keep in mind when filming
- You can speak before doing the work or while doing it. Action subtitles are timed to when the movement is visible in the picture, not to when you spoke.
- In places with loud machine noise, bring the microphone close to your mouth. Parts that cannot be heard will not become subtitles.
- If there is a lot of conversation unrelated to the work, the changes in the talk cannot be used as clues for step splits.
- Match the language you speak with the "Recognition language" in "Advanced settings" of the AI processing options.
- For a video recorded with no narration at all, check "Videos without narration (analyzed only by video)" in the AI processing options. This prevents step splits from being mechanically divided into equal time segments.
If you say a cue such as "Caution here" while filming, you can also keep questions only for the steps where those words were spoken. For details, see Keep questions only for steps where a keyword was spoken (narrowing down questions).
5. Choose which subtitles to use when creating an SOP
When you create an SOP from material that has finished AI processing, choose the subtitles to include in the SOP under "Subtitle types to adopt" in "Subtitles" of "Step conversion settings."
| Option | Content of subtitles | Suitable videos |
|---|---|---|
| AI-generated (ON by default) | Description of actions read from the picture | Videos of work being filmed, videos with little narration |
| Speech transcription (OFF by default) | What the filmer or worker said | Videos filmed while talking, videos where lectures or explanations are the main focus |
| Both | One subtitle track combining speech and action subtitles along the timeline | Videos where you want to show both actions and explanations |
If you want what was said while filming to appear as subtitles, check "Speech transcription." Your selection is remembered and becomes the default the next time you open Step conversion settings. Subtitle text is searchable, so if you include speech subtitles, you can later find the relevant section by searching for words.
The subtitle type is chosen when you create the SOP. If you want to switch all subtitles to the other type after creation, create the SOP again from the same material. For details, see Step conversion settings (read-aloud, subtitles, questions, related files).
Fixing analysis results
- You can fix subtitle wording in the video editor (Editing subtitles)
- If the same term is repeatedly misheard, register that term in the Terminology Glossary, then re-analyze it from "Check AI analysis results" → "Re-run AI." Later analyses will also write it in that spelling.
- If you want to compare results under different conditions, copy the material and then re-analyze it (Try different analysis settings (copy the material and re-analyze))
Frequently asked questions
Q. Can I create an SOP if I don't say anything while filming?
A. Yes. Action subtitles, step splits, and body text are created from the picture. For a video recorded with no narration at all, check "Videos without narration (analyzed only by video)" and then analyze it.
Q. What I said does not appear as subtitles.
A. "Subtitle types to adopt" in "Step conversion settings" may be set only to the default "AI-generated." Check "Speech transcription" and then create the SOP again from the same material. If the material has no speech data, "(No data)" is displayed and cannot be selected.
Q. Is there any effect if I only say single words?
A. Single words do not connect what you are doing or how, so they become less useful as clues for step splits or body text. Please speak a short sentence that matches the work, such as "Next, remove XXX."