Now it gets concrete. Every recommendation in this part comes with its reasoning from Part 3 — and with a principle that belongs written down here explicitly: the pictures are made by the person behind the camera. What follows are the settings with which the reconstruction succeeds most reliably. Anyone who knows them and decides otherwise for creative reasons makes no mistake — they make an informed choice. That is exactly what the reasoning is there for.
AI image
Autofocus is a blessing in everyday use and a quiet saboteur in series work: it refocuses before every image — now on the front edge, now on the middle, now on the background. With that the plane of sharpness jumps through the series, and with it even the “window” from Part 3.4 changes slightly (the effective focal length shifts a little when focusing — photographers know it as focus breathing). Both are exactly the inconsistencies that disturb the camera computation.
The handling is simple: set sharp cleanly once — by autofocus, that is allowed — then switch to manual focus and do not touch it again for the entire series. When doing so, do not focus on the frontmost edge but on roughly the front third of the object: the depth of field extends further behind the focus plane than in front of it. And keep the distance to the object as constant as possible during the circuit — then the locked sharpness stays valid everywhere.
Autofocus refocuses before every image — once locked, the plane of sharpness stays exactly the same for the whole series.
The stabiliser compensates for hand tremor by moving the sensor or lens group in the opposite direction. Handheld this is pure gain: leave it switched on. On the tripod the logic reverses — the stabiliser searches for movement that is not there and can itself drift or oscillate minimally. The result is mysteriously soft images despite the tripod. Hence the double rule without exception: handheld on, on the tripod off. (Some modern systems detect the tripod themselves — whoever wants to rely on that checks it on a test shot at 100-percent view.)
From Part 3.3 you know the goal: the whole object sharp, from the frontmost to the rearmost edge. The tool for it is the aperture — and its optimum lies in the middle. Too wide open (f/1.8, f/2.8) and the sharpness becomes a narrow disc; too far closed (f/16, f/22) and diffraction eats the detail sharpness back up — not dramatically, but measurably. As a starting point for close-object work: f/8 to f/11 on full frame, f/4 to f/5.6 on Micro Four Thirds — that is the same equivalent aperture, only translated per sensor (the small sensor needs two stops less for the same depth, its advantage from Part 5.1).
The practical test beats any table: take a test shot and check at 100-percent view the frontmost and the rearmost edge of the object. Both sharp? The aperture fits. Soft at the back? Close one stop or step back a little. Motion Guard, by the way, also assesses the aperture of your series — and does so sensor-aware by exactly this logic: the middle wins.
Too open: sharpness becomes a narrow disc. Too closed: diffraction eats detail sharpness again. Between the two lies the royal road.
Photographers usually learn the exposure triangle with the priority “ISO as low as possible”. For the reconstruction it is re-sorted, and the order is a small provocation: first the aperture (it secures the depth — non-negotiable), second the shutter speed (it prevents camera shake — blur is incurable), third, last, the ISO. Why of all things ISO last? Because moderate noise is by far the smallest evil: it is randomly distributed differently in every image and largely averages out over many images in the training — whereas blur and shallow depth are the same, irreversible lie in every image. Better ISO 1600 and sharp than ISO 200 and shaken.
For the safe handheld speed the old rule from Part 1 of the Motion Guard world applies: shorter than 1 divided by (focal length × crop factor) — at 25 mm on Micro Four Thirds therefore shorter than 1/50 second; whoever wants a reserve halves it once more. The tripod dissolves the triangle entirely: base ISO, desired aperture, any speed — then stabiliser off (6.2) and release via self-timer or remote so that the pressing itself does not wobble. And the consistency rule from Part 3 applies here too: set the exposure fixed (mode M), so the automatic does not jump in the middle of the series.
Here, on purpose, there is no recommendation table. The focal length is a creative decision — it determines angle of view, working distance and perspective effect, and it belongs to the person behind the camera. As far as the reconstruction is concerned, only two physical truths count: extremes have side effects (very wide distorts and makes the object small, very long demands distance and amplifies camera shake — Part 5.4), and within a series the focal length stays constant (Part 3.4). The zoom is a wonderful tool between series — within the series it is taboo. Whoever seeks a good starting point without thinking long: the normal range around 35–60 mm full-frame equivalent is low in distortion, handy and fit for everyday use. But it is a starting point, not a commandment.
RAW is the first choice, where available: more reserves in shadows and highlights, lossless white balance, and modern converters pull visibly more usable detail from a series with denoising and sharpening — that is allowed and desired, as long as the same settings act on the whole series. A good camera JPG is no disaster; but whoever develops anyway develops serially. If the camera shoots double formats (RAW+JPG or special formats such as high-resolution supplementary files), the rule is: import only one of the two — otherwise the same vantage point counts twice.
The three development prohibitions all follow from Parts 3.4 and 3.5: no geometry (cropping, rotation, perspective correction — they destroy the window), no single-image outliers (pulling one image brighter than the others = a witness contradiction), and no retouching in single images (the clone stamp removes the spot only in one image — for the reconstruction that is a witness who lies; distractions are removed on the finished 3D model, not in the source images). On export: full resolution, high quality, sRGB.
After the import DFX SplatCore checks every series automatically: shake risk from the capture data, actual sharpness from the pixels, both weighed against each other — including the tripod detection from Part 5.5 and the correct crop factor for your camera model. The result appears as a finding in the telemetry; the individual verdicts per image stand in the log. And the most important sentence about it, for the third time, because it is the foundation of trust: nothing is sorted out. Motion Guard reports — you decide.
This is how the checker becomes a tool: before important appointments send a test series through the import and read the finding — that is a free dress rehearsal for light, aperture and speed, before it counts. The details of both check paths and of the device database stand in the two Motion Guard manuals of this library.
