Use the panel containing the actual performance, not the whole tutorial layout. First identify what each panel represents, then select a complete time segment and crop that panel into a separate file. Keep the original recording and inspect the exported crop before treating it as motion evidence. Use material you own or have permission to reuse.
Identify the source before choosing a crop
A tutorial may show the original actor, a processed control representation and a generated result simultaneously. Similar timing does not make those panels interchangeable.
Play the sequence once while watching only one panel, then repeat for the others. Look for labels and an explanation from the creator. Decide which panel demonstrates the movement you intend to drive. If you cannot establish that role, do not guess solely because one version looks sharper.
The finished AI result is useful for understanding the tutorial's goal. It is not automatically the input that produced it.
Separate time selection from picture selection
Choose the time interval first: include the preparation, main action and recovery. Then select the rectangular picture area containing that entire performance throughout the interval.
Record both decisions in your notes. “Top panel, from the neutral stance through the return” is more useful than “the dance section.” A panel can change size or switch to another shot halfway through a tutorial, so scrub the entire chosen interval.
Premiere documents crop and position/scale controls for selected clips. Those editing controls prepare a file; they do not certify that the selected panel is a valid input for your generation model.
Handle captions without sacrificing the action
| What covers the panel? | Sensible response |
|---|---|
| Label outside the performer area | Crop it away if the full action still fits |
| Caption across moving hands | Find a clean original or another permitted source |
| Side-by-side result comparison | Export only the identified performance panel |
| Text appearing halfway through | Review every affected frame before keeping the segment |
| Missing feet below the panel edge | Choose a different source for foot-dependent movement |
A tighter crop is not necessarily cleaner evidence. If removing a label also removes the raised hand, the reference has lost the event you need. Do not fill that area with generated anatomy and present it as the original recording.
If no clean version exists, decide whether the remaining visible movement is sufficient for a limited test. State the limitation instead of expecting automatic overlay removal.
Audit the exported reference
Open the new file outside the editor. Check that only one panel remains, the performer stays visible and playback begins where expected. Confirm that the crop did not stretch the body to fill a different frame shape.
Compare the saved first and last moments with the tutorial. If the action begins midway through a turn or ends during a recovery, adjust the interval. Keep the original speed for the first motion test so that timing differences remain easy to diagnose.
Name the file with its source role and excerpt rather than calling every export “final.”
Two questions during preparation
Can I upload all three panels and explain them in text?
Use that arrangement only if the selected tool explicitly supports it. A written explanation does not guarantee that a single-motion input can separate three simultaneous performances.
Will cropping remove text that crosses the person?
Only when the text lies outside the retained rectangle. Text over the subject remains part of the selected frames; finding a cleaner source is usually the more informative preparation step.