Ctrl-Alt-Win

Stepwise: record a task once, get a written SOP, and catch skipped steps later The problem Factories, labs, warehouses and repair shops depend on Standard Operating Procedures (SOPs), written step-by-step instructions for doing a task correctly. Writing them is slow and manual: someone performs the task, takes photos, types up each step, and formats a document. Many SOPs are outdated or never written at all, so new workers learn by watching, and steps get skipped, including safety steps. What Stepwise does An experienced worker records themselves doing the task once and says each step out loud as they go. Stepwise turns that one video into an illustrated SOP: numbered steps, each with a title, clear instructions, a way to check it’s done, any warnings, and a photo from the video with the key parts highlighted. How it works Splitting the video into steps. Stepwise samples the video at 2 frames per second and measures how much the image changes between frames, to find when the hands are working and when they pause. Muse Voice Transcribe turns the narration into sentences with timestamps. Merging speech and motion. Each spoken sentence is matched to the burst of activity it belongs to. People often talk before or after they act, and sometimes do two steps without pausing, so when two sentences fall into one burst, Stepwise splits it at the moment the hands slowed down most. On the test video, motion alone found 5 of 6 steps; combined with speech, all 6 were correct. Picking keyframes. For each step, Stepwise chooses the clearest frame from the middle of the action, scored on sharpness, stillness and how close it is to the middle of the step. Muse Spark reads the narration and the photo for every step and writes the SOP text. It’s told to trust the photo when the narration is late or vague, and it names the objects in each photo that matter for that step. SAM 3.1 takes the object names from Spark, finds those objects in the photo, and Stepwise draws labeled boxes on them. The finished SOP is exported as a single web page with the images built in, so it can be opened or shared anywhere. Why the Meta models are used this way Each model does what it’s best at: Voice Transcribe hears, Muse Spark reasons and writes, and SAM 3.1 sees exactly where things are. Spark decides what matters in each photo, and SAM finds where it is. Around them, Stepwise’s own code handles timing, step splitting and image selection. Every API response is cached, so each video is processed only once and repeated runs cost nothing. Who it helps Operations and training teams that need SOPs quickly, small manufacturers and workshops without dedicated technical writers, and labs that need consistent procedures. A second mode is planned: compare a new recording against the SOP and flag skipped or out-of-order steps, for training and quality checks.