DE-instruct creates the instructions from what is said and shown in the video. Anything missing from the video won’t be in the instructions either.
“Now enter the batch number: year, month, day, then the batch.”
The key answers up front. The details follow on the next slides.
As long as the task takes. It’s best to split several tasks into separate videos. The longer the video, the longer processing takes and the more credits it uses.
Yes. Long pauses and searching for things can be cut out. But don’t insert any text slides or title cards, or they will end up as step images in the instructions.
Not recommended. Without speech, the AI is missing its most important source. The instructions may then have gaps or contain steps that didn’t happen that way.
For minor slips of the tongue, “Correction: it should be …” is enough. If a hand movement was shown incorrectly, it’s better to record that part again.
They’re a draft. Check them against the video, add any missing details, and only then publish them.
Good instructions start before the shoot. Ten minutes of preparation saves a lot of rework later.
One video, one task, one variant.
Run through it without the camera.
Target values and numbers on a sheet of paper.
Camera mounted, clip-on microphone on.
Wear PPE as required.
The AI discards blurry images. The step then has no photo in the instructions.
Stick to this order. Each part fills one section of the instructions.
The audio is the most important source. Each sentence fills one field in the instructions, and the AI has to guess whatever is missing.
“Right, now the lid goes on … yep, this way round … that’s it.”
The AI doesn’t see the whole video, only individual still frames from it. Anything shown only briefly or partly hidden gets lost.
Tap from below or from the side. Also read the value aloud.
Hold the camera steady, keep hands and part in focus.
Otherwise, inserted slides end up as step images in the instructions.
Show and name tools and parts once.
Straight on, no glare, read the value aloud.
Wear it and mention it as well.
Check these off right before the shoot.
Talk through one step, just as you would when filming. The coach checks it against the five sentences and shows what’s still missing.
The recording is sent to the AI only for this analysis and is not stored by this tool.
Check a photo, a screenshot or a short video (up to 10 seconds). The coach shows what the AI can recognize and read in it, and what you can do better when filming.
The image is downsized in your browser and sent to the AI without location data. This tool does not store it.