Videos

The Videos tab is where you add the videos you want to analyze. Write instructions in plain language, run the analysis and review the output.

Adding Videos

Before the AI Watcher can analyze anything, it needs videos to work with.

Click Add Videos to add videos from your library. Alternatively, go to Archive, select one or more videos, and use the Bulk Actions menu to add them to a workflow from there.

Your videos appear in the Workflow Videos table. Each row shows the video name, its current status, a score once it has been watched, any tags or notes you have added, and when it was created and reviewed.

Tip: Use Metadata Columns to add custom fields to the table, and Show Filters to narrow down large video lists by status, tags, or date.

Writing your AI Watcher Instructions

The AI Watcher Instructions box is right below the video table. This is where you tell the AI Watcher what to look for and what to return.

Click Edit Instructions to open the field and start writing.

Good instructions specify the goal, the format, and any constraints. The more specific you are, the better the output. Here are some examples:

• “Watch this soccer match and identify highlight moments for both broadcast and social distribution. For broadcast, include the most important plays and reactions with timestamps. For social, identify single-subject moments with centered framing.”
• “Extract every mention of pricing or contract terms. For each one, include a direct quote and a timestamp.”
• “Flag any statements that could be a compliance risk. Include the exact quote and timestamp for each one.”

Three optional settings sit below the instructions field:

• Enforce JSON Output: Returns results as structured JSON. Use this when the output needs to feed into another tool or system.
• Run on Chunks (Beta): Processes the video in segments rather than all at once. Useful for long videos or content-dense footage.
• Use Transcription (Beta): Incorporates spoken audio into the analysis. Best for dialogue-heavy or interview footage.

Click Use Template to start from a pre-built instruction set and edit from there. For more guidance on writing effective instructions, see the Prompt Guide.

Running the Workflow

Once your videos and instructions are in place, click Watch All Videos.

Flowstate sends your instructions to the AI Watcher, which begins processing every video in the workflow. A confirmation banner appears at the top of the screen, and each video’s status updates to In Progress as the Watcher works through them.

When a video is finished its status changes to Watched and a Score appears in the table. The Score reflects how well the output matched your instructions, rated out of 5.

Alternatively, click the three dots on the right side of any video row to open the Actions menu. From there you can click Watch if you only want to watch a specific video in the workflow.

Tip: You can download the output for all videos at once using Download JSON in the top right of the Videos tab. This is useful when the output needs to feed a downstream system or spreadsheet.

Reviewing Output

Click any video name in the table to open its output page. This is where you review what the Watcher found, validate the results, and turn moments into finished content.

The output page has six sections:

AI Watcher Output is the results generated from your instructions. Depending on what you asked for, this may be a list of timestamped moments, structured clip recommendations, extracted quotes, or flagged issues.

Each result includes a clickable timestamp that jumps the video player to that exact moment for quick verification.

You can copy the full output, or click Edit to refine the wording before using it elsewhere.

AI Watcher Score is a score out of 5 that reflects how well the Watcher performed on that video. Below the score, you will see two sections:

• Analysis: What the Watcher did well, such as following the prompt structure or accurately capturing key moments.
• Improvements: Specific suggestions for refining your instructions to get better results, such as “instruct the model to verify player names against transcript roles.”

Use the Improvements section to refine your instructions and improve output quality across the batch.

Transcript is a timestamped record of everything spoken in the video, synced with the video player. Click a timestamp to jump to that moment. Use the search field to find specific words or phrases and jump directly to where they appear.

Video Details shows the tags and notes attached to this video, along with a Reviewed toggle you can check once you have looked over the output. Use this to track which videos in a large batch have been reviewed.

Publish History shows whether this video has been published and when. If it has not been published yet, this section will say so.

Edit History is a record of any edits made to the output. If your team has refined or updated the Watcher output, those changes are logged here so nothing is lost.