When you create an SOP from a video or file, you can specify how to analyze it using "AI processing options". The SOP will be created even if you press "Run" with the default settings. This section explains what each item does and how to choose them.
AI processing options are available in all plans. The number of times you can run AI analysis varies by plan.
Open AI processing options
The same screen opens with any of the following:
- It displays automatically when you upload a video or file
- For materials that haven't been analyzed, open "Run AI processing" from the material details modal (see Run AI analysis on unanalyzed video materials)
- When re-running analyzed materials, open "Re-run AI" from "Check AI analysis results" (see How to re-run failed AI processing)
The screen is divided into three sections: "Video description", "Analysis options", and "Advanced settings". You usually adjust the first two; "Advanced settings" can remain at defaults.
Video description
This is where you tell the AI what work the video shows. Mentioning the equipment involved, the purpose of the work, the tools used, and so on improves analysis accuracy.
You only need to describe what is unique to this video. Common setup like in-house terminology and equipment names — any reference that applies to many videos — can be registered once in the Terminology Glossary, and the AI will use it in future analyses without re-entering.
You can also insert a saved template by clicking the icon at the top right of the field.
Analysis options
Split granularity
Specify how finely you want the video divided into steps.
| Choice | Split unit |
|---|---|
| Process | Split into large units such as setup, execution, and cleanup |
| Step | Split at standard work step units. This is recommended when publishing as an SOP |
| Action | Split finely at individual action levels such as button presses and tool usage |
| Single | Do not split; treat the entire video as one step |
Default is "Step". Choose "Single" when you want to use it as a single edited video without step splits (see How to create a single edited video without step splitting).
Text style
Specify the voice and expression used in the step body text.
| Choice | Content |
|---|---|
| Command | Plain form and imperative. The standard for SOPs |
| Formal | Polite language. For new employee training or SOPs submitted to business partners or audits |
| Suggestion | Soft, conversational tone. For training and education |
Default is "Command". If you have registered "How to write SOPs", the option "Follow policy" is added and becomes the default.
Subtitle style
Specify the voice and expression of the action subtitles shown in the video. This can be set independently from the body text, with the same three options as text style plus "Match text". Default is "Command".
When you choose "Match text", the subtitles follow the body text style. However, when the text style is "Follow policy", you cannot choose "Match text", so please select the subtitle style individually.
Videos without narration (analyzed only by video)
When checked, audio is ignored and only the video is analyzed. Step splits and action descriptions are generated as before, but speech transcription is not.
Use this for videos without narration, videos with only background music or machine sounds, or videos where speech is unclear. It's also effective when re-running videos where audio caused AI analysis to fail.
If you analyze a video without narration without this setting, the step split may become mechanically divided by time rather than by work boundaries. The AI looks for narration topic changes as one clue to boundaries; when speech is unrelated to the work, the AI loses that clue and instead pursues only the target step granularity.
When analysis results show this state, a notification saying "The step splits look mechanical" appears just before creating the SOP.
- Creation doesn't stop. For work that progresses in a steady rhythm, equal splitting may match reality, so you can choose "Create as is"
- Clicking "Re-analyze" opens the AI analysis options dialog with "Videos without narration (analyzed only by video)" visible. Decide whether to check this option yourself
- Re-analysis from this notification does not count toward your AI processing run quota
Create questions (when creating from a file)
Appears when creating an SOP from a file (PDF or Office file). When checked, the AI creates questions for workers based on the material content. The default is ON.
When OFF, only questions already written in the original material are imported (legacy behavior). The app doesn't create questions for cover pages, tables of contents, revision histories, and other pages without work content. Questions created can be removed later in "Step conversion settings" using "Apply questions to steps".
When "Body writing style" is "Transcription", it's in transcription mode so this option is not available. In this case, only questions in the original material are imported.
Question type
Specify how the AI creates questions. For video-based SOPs this is always shown; for file-based SOPs it appears only when "Create questions" is checked.
| Choice | Content |
| On-site check | Create questions for workers to verify their own work on the job. The answer options show only things that can realistically happen in that process; unrealistic wrong answers are not included. You can use this as a digital version of pointing and calling |
| Comprehension test | Includes unrealistic wrong answers to test memory. Choose this when using for training or proficiency confirmation |
Default is "On-site check". The SOP player assigns no score or pass/fail and requires you to answer before proceeding, so the default is set to avoid workers getting stuck mid-procedure because they can't remember.
The setting you choose here carries over to the question settings when you create the SOP. Materials analyzed with "On-site check" have "Continue when unexpected" turned ON by default in step conversion settings.
Advanced settings
Target step count
Specify the target number of steps. Default is "Auto", where the AI judges fineness based on split granularity.
| Choice | Content |
|---|---|
| Auto | AI judges based on split granularity (default) |
| Specify step count | Specify roughly how many steps to divide the entire video into |
| Specify seconds | Specify roughly how many seconds per step |
Even when specified, how steps are split (dividing at semantic work boundaries) doesn't change. The target applies only to the number of steps, and it's a target, not a guarantee of that exact number.
When split granularity is "Single", this option doesn't appear since it would conflict with the "one step for entire video" setting. When using "Reference work SOP" to match resource boundaries, the resource step count takes priority so this isn't used.
Recognition language
The language of the speaker in the video. Setting it correctly improves transcription accuracy. Even if a language isn't in the list, you can type it in the text field and search for it.
Aggressively cut sections without speech
Broadly exclude sections without narration from the created SOP. Default is OFF.
Even when OFF, obvious waste like filler and long silences is cut conservatively, but natural pauses during work are kept. When ON for a video where work continues silently, the work itself may be cut, so the default is OFF.
Turn ON when using a narrated video or a lecture-style video with continuous speech, and you want to compress any dragging sections. Cut sections can be restored in the video editor.
If "Videos without narration (analyzed only by video)" is checked, this option is unavailable since narration is not assumed.
Reference work SOP (Beta)
Specify one existing SOP file (PDF, Excel, Word, or image) to divide the video to match that resource's splits, step count, and order. Use this when you want to carry forward the structure and terminology of an already-created SOP into your video manual.
When specified, split granularity and target step count are not used. For details, see Specify a reference SOP in video analysis (Beta).
When to use split granularity vs. target step count
| What you want | Setting |
|---|---|
| Create an SOP and let AI decide | Split granularity = Step, Target step count = Auto |
| Want finer or coarser splits | Change split granularity to Action or Process |
| Want to match step count | Specify the target step count |
| Want to match existing SOP boundaries | Specify a reference work SOP |
| Don't want to split into steps | Split granularity = Single |
When creating an SOP from a file
When creating an SOP from a file (PDF, Excel, Word, or PowerPoint), the options differ from video. Split granularity, subtitle style, target step count, and reference work SOP do not appear.
| Item | Content |
|---|---|
| Conversion method | Page layout setting for Office files. Choose from "1 sheet = 1 page", "Fit width to 1 page", or "Print settings". If the original file's print settings are correct, "Print settings" is recommended |
| Step split method | How many steps per page. "Per content" divides table rows and bullet lists within a page into separate steps. "Per page" makes one page one step. In both cases, each step stays within one page and doesn't span pages |
| Body writing style | "Formatted" reads the content and organizes it into work content, tips, cautions, and safety. "Transcription" copies the original text as-is. Use "Transcription" for materials like regulations and checklists where the original wording must be preserved |
| Create questions | The AI creates questions for workers based on material content. Default is ON. When OFF, only questions already in the original material are imported (legacy behavior). Not available when body writing style is "Transcription" |
| Question type | Appears only when "Create questions" is checked. Content is the same as for video (On-site check or Comprehension test) |
For details, see Create an SOP from a file (Excel, Word, PowerPoint supported).
Saved settings
Settings you specify are saved and become the defaults the next time you open these options. When you "Re-run AI" on analyzed material, conditions from the previous analysis (reference work SOP, target step count, etc.) are restored.
You can check what conditions were used to analyze material by viewing its "Check AI analysis results".
Common questions
Q. How long does analysis take?
A. Roughly the same as the video length (about 10 minutes for a 10-minute video). Analysis continues even if you close the app, and you'll receive an email when it's done. You can analyze up to one hour of video per job.
Q. Is there an option to not generate subtitles?
A. No. Subtitles are always generated. If you don't want speech transcription, check "Videos without narration (analyzed only by video)".
Q. Can I change options and re-analyze?
A. Yes. From "Check AI analysis results", click "Re-run AI" to change options and re-analyze. The previous results are overwritten.
Q. The step splits don't match what I expected.
A. First, try changing the split granularity and re-analyzing. To match a specific step count, specify the target step count. To match an existing SOP structure, specify a reference work SOP. If still not right, use the video editor to manually edit steps.
Related information
- Step 1: Upload a video
- Run AI analysis on unanalyzed video materials (run AI processing later)
- Specify a reference SOP in video analysis (Beta)
- Define how to write SOPs (reflect your organization's style and standards to AI)
- Terminology Glossary (terms and multilingual translations)
- Unify notation inconsistencies (Notation Unification Dictionary)
- How to re-run failed AI processing