Editing on the Desktop: Keyframes and Easing, Optical-Flow Slow Motion, Log Grading by Waveform, Auto-Cutout, Masking and Motion Tracking

A keyframe locks a setting at a point in time so it can change over time - any parameter with a diamond icon qualifies: position, scale, colour, opacity. The everyday use is a punch-in, keyframing scale and position together so the frame tightens as a subject does something. The detail that separates amateur from professional is what happens between the keyframes: by default the change is linear, which reads as mechanical, and selecting a keyframe and applying auto-curve eases it into something that looks intentional. This is the same easing point the editor-hiring critique in the AI Creator Course marks people down for missing.

The floor is arithmetic, and below it the clip stutters because the frames do not exist - 0.1x on 60fps footage shows every held frame. When you need to go further, the software invents frames, and the two methods trade differently. Frame blending is fast and leaves visible ghosting. Optical flow takes much longer to render and is far more convincing, though fast-moving edges - he demonstrates a dog's paw - still warp. He notes that he is 'surprised they included that level of power in the free version of capcut'. LumaFusion caps at about 0.4x and offers a maintain-audio-pitch toggle alongside.

Log footage is deliberately flat and has to have its contrast put back. Doing it by eye on an uncalibrated screen is guesswork; doing it on a waveform is not. Turn the oscilloscope on, then in Curves pull the shadow point down and right and the highlight point up until the data reaches both edges of the graph - and stop there, because going past either line crushes blacks or clips highlights and destroys detail permanently. Then return to the basic panel and add back the saturation the log profile removed. The same three steps worked unchanged on Sony S-Log3 and on Apple Log from an iPhone. A LUT remains the one-click alternative when you have one matched to the profile.

Duplicate the clip onto a layer above, apply auto cutout to isolate the subject, and put the text on a layer between the cutout and the original - the words now sit behind the person. Auto cutout struggles with motion blur, so a mask over the cutout layer can reveal only the clean area and hide a glitching hand. The same masking idea builds the circular floating-head picture-in-picture used over screen recordings, which is, he notes, exactly how the course's own lessons are made: record the screen and the face separately, mask the face to a circle. A mirrored horizontal mask across two takes of a tripod-locked shot gives the clone effect.

Trackers fail on flat, low-contrast areas because there is nothing to lock onto, so the choice of tracking region decides the result before you press anything. Start the track at the very beginning of the clip, box a distinctive high-contrast feature, and - the detail most people miss - turn scale and distance tracking off, keeping position only, which usually looks more professional and avoids the graphic breathing. Give the attached graphic an in and out animation rather than letting it pop. Tracking also works from partway through a shot, demonstrated on a bus entering frame.

The order is record under peak, because clipped audio 'can't really save', then apply loudness normalisation, which lifts the quiet passages and pulls the loud ones down toward a broadcast-like consistency, then fine-tune by hand while watching the meter stay out of the red. The target band is -12 to -6dB. Noise reduction is available and useful for wind or ambient hum, but he is careful not to oversell it: 'it's not gonna totally save terrible audio.' The same numbers appear in the Day 7 audio training, which is a good sign they are the house standard rather than an app default.

The desktop stabiliser offers the same ladder as the mobile one - minimal cut through to most stable - and the same trade: more crop, more stability, more risk of the frame edges melting. His rule is unambiguous, because a viewer forgives a little movement and does not forgive warping: 'having that kind of jello look to your video is just very unprofessional.' Default to the minimal or recommended settings and step up only if the shot is otherwise unusable. LumaFusion's equivalent is Lock and Load by CoreMelt, which analyses motion first and then exposes a strength control.