This release focuses on improvements to video AI analysis. In addition to having AI split videos automatically, you can now specify the desired granularity of steps as a target and align splitting with the structure of an existing SOP, so analysis adapts to your needs on the floor. We've also improved splitting accuracy for longer videos and the overall experience of re-analyzing.

① Added "Target step count" to AI processing Options (Video)

Previously, step granularity could only be specified using [Split granularity] (Process / Step / Action / Single), making it impossible to make adjustments like "I need finer splits" or "I want this video to break down to roughly 10 steps total." We've now added [Target step count] to [Advanced settings] in AI processing Options.

• Auto (Default): AI judges fineness based on Split granularity, as before
• Specify step count: Set approximately how many steps the entire video should be divided into (example: 10 steps)
• Specify seconds: Set the target length per step (example: 30 seconds/step)

Even when specified, how steps are divided (at meaningful work boundaries) doesn't change. Only the number is adjusted to match your target. Long videos are analyzed in segments internally, but adjustment is made across the overall video to meet your target, so step counts won't vary based on video length.

● Caution

• If you don't change the setting, analysis results are the same as before (Auto is the default)
• This option is not shown when Split granularity is set to [Single] (treating the entire video as one step)
• When you align with the reference material's boundaries using [Reference SOP] (described below), the material's step count takes priority and Target step count is not used (the screen shows it as Disabled)
• This is a target; it doesn't guarantee the split will be exactly that number

● AI processing Options > Advanced settings > Target step count

② Improved splitting for Reference SOP (beta)

We've improved the feature (in beta) that lets you specify an existing SOP (PDF / Excel / Word / Image) as a [Reference SOP] and split your video to match that material's structure.

• Long videos now match the material's step count: Previously, splits would misalign in longer videos, resulting in a different number of steps in the final SOP than in the reference material. Now the entire video is aligned to match the reference material's step count and order
• Steps missing from video no longer drop out: Steps written in the material but not filmed would disappear from the SOP. Now we retain all material steps while mapping video content to the steps that have footage
• Clearer checkbox label: [Align video split to this material's boundaries (strict alignment)] → [Split according to the document's steps]

For usage details, see Specify a reference SOP in video analysis (beta). Since this is still in beta, please send any feedback—whether it worked well or didn't split as intended—to customer-success@episotech.com (Dive Customer Success).

③ Previous analysis settings now carry over when re-running

When doing [Re-run AI] on materials or retrying only failed sections, previously specified conditions weren't carried over, and intended settings wouldn't apply. We now preserve the following settings when re-analyzing:

• Reference SOP and how it's used (Split according to the document's steps / Use as reference for terminology and context only)
• Target step count
• Silent cut settings

Even when retrying only certain sections in a long video, they're analyzed under the same conditions as other sections, so you won't see inconsistencies in granularity or tone between sections.

Related information

• AI processing Options items and how to choose them
• Specify a reference SOP in video analysis (beta)
• Run AI analysis on materials that haven't been analyzed (apply AI processing afterward)
• How to re-run failed AI processing

If you have any questions about this update, please feel free to contact customer-success@episotech.com.