We have significantly expanded the Work Analysis feature and released two new steps: "Compare" (side-by-side view of target person's work video and standard) and "Extract" (compose a model cycle from multiple cycles). We are also implementing cycle analysis support (repeat type declaration / official support for single-cycle videos / AI model enhancement / stabilization). This article introduces the details of this update.

① Work Analysis "Compare" Function — Visualize skill level progression with the target person

Using the confirmed standard (after-improvement video + element definitions) as the baseline, the system now uses AI to watch and align the target person's (junior staff, new hires, etc.) repeated work videos, making it possible to visualize skill level progression through cycle time comparison, delay per element, and AI coaching.

• Previously: Trainers could only reflect subjectively, saying "it seems much faster" or "this part seems slow," without being able to discuss objective numbers
• Now: Display the standard and target person's videos side by side with synchronized playback, and visualize delays per element with red shading. You can instantly see "You are delayed by ○ seconds in element ○○; moving this way will be faster"

●Main Features

• Select target person and video for one-click comparison: Choose the target person from the team members list and select their work video from the material library to start
• Cycle time comparison: Display each cycle (standard + target person) as stacked bar charts side by side, including average and standard value lines
• Delay per element: Show the difference from standard = 0 as a line chart, with the selected cycle as a solid line and average as a dotted line
• Variance per element: When you have two or more cycles, display shortest – longest range + average + selected cycle + standard value on one screen
• AI coaching: AI generates individual improvement suggestions per element based on the difference between standard and target person (with priority, element, points, and advice)
• Synchronized playback of standard + target person: Compare videos displayed side by side at a common playback position
• Cycle adjustment and re-analyze: If AI's automatic segmentation is off, you can fine-tune it. If unsatisfied, run re-analyze
• Re-analyze failed chunks only: If part of a long video fails, re-run only that section (does not consume AI count)

For instructions, see STEP6 Compare: Compare target person's work video with standard (skill level visualization).

② "Extract" Step — Compose one model cycle from multiple cycles

From a video showing repeated work, the system visualizes cycle variance and adds a new step to assemble the "model cycle" to follow. Classification and improvement that follow are performed against this model.

• Previously: Repeat videos could only be analyzed "in general overall," making it difficult to combine best elements and create an ideal form
• Now: An element-by-cycle matrix displays cycle time for each element, and you can compose the model by simply clicking cells where green = fastest / red = slowest / blue ✓ = currently selected

●Main Features

• Cycle analysis: AI automatically extracts cycles and element work from multi-cycle videos
• Two presets:
  • [Best per element] = Bulk-adopt reliable best iterations for each element (composite of element bests)
  • [Best single cycle] = Use the best cycle overall with all elements aligned (consistent demonstration)
• Manual cell-by-cell override: After applying a preset, you can also select individual elements from different cycles
• Model continuous preview: Play selected elements in sequence to confirm the model's appearance on the spot
• Cycle adjustment: Fine-tune boundaries if AI's automatic segmentation is off
• Single-cycle videos proceed directly to the next step: Since a single-cycle video does not require composition, you can confirm and proceed directly to classification

For instructions, see STEP2 Extract: Extract element work and create one model cycle.

③ Cycle Analysis Support (Enhanced accuracy and stability)

To make cycle analysis that supports the "Extract" step even more user-friendly, we have significantly strengthened the underlying foundation.

●Declare "repeat presence" during analysis

During AI analysis of element work, the system now asks the user: "Does this video repeat the same work?" By passing the answer as a hint to the analysis, extraction accuracy has improved.

• Yes: Analyze as multiple cycles, and in the model step you can align each cycle to compose the model
• No: Analyze as a one-time task and proceed directly to classification and improvement

●Single-cycle videos can now be analyzed directly

Previously, analysis assumed "repeated work," so results were sometimes unstable for videos containing only one cycle. With this improvement, single-cycle videos can now be correctly analyzed as "one-time tasks". After confirmation in STEP2 (Extract), proceed directly to classification.

Step Structure Update (4 steps → 6 steps)

With the addition of new features, the Work Analysis editor has been expanded to 6 steps.

1. STEP1 Selection: Select the original video and review AI improvement suggestions
2. STEP2 Extract (new): Extract element work + compose model cycle
3. STEP3 Classification (former STEP2): Classify in three categories
4. STEP4 Improvement (former STEP3): Cut, rearrange, and speed adjustment
5. STEP5 Review (former STEP4): Before/after comparison preview + export
6. STEP6 Compare (new): Compare target person's work video with standard

For the full picture, see Work Analysis overview and getting started.

We look forward to your continued use of Dive ◎