A photograph of a cracked weld taken from arm's length tells the reader almost nothing. The first frame should place the part: the whole component, or the whole section, on a clean floor with something for scale beside it. Only after that does a close frame of the failure make sense, because a close frame with no context cannot be located.
Send both. A wide frame answers where, and a close frame answers what. Claims that arrive with only close frames are the ones that need a second round of messages, and a second round costs days nobody has.
Raking light is the trick worth learning. Put a torch low and to one side of the joint so the beam skims across it rather than into it. Cracks, laps and undercut all cast a small shadow that a head on flash flattens out completely.
Daylight through a workshop door is usually better than a phone flash. Flash reflects off polished titanium and polished stainless and turns the area you care about into a white patch. If the part is brushed or polished, angle it so the reflection falls away from the joint rather than across it.
Most genuine faults in welded work are at a joint: the toe of a weld, the edge of a flange, the root where two tubes meet. Photograph the toe line along its length rather than head on. A crack that starts at the toe and runs into the parent metal looks very different from a crack that runs through the weld throat.
Where a failure is inside a tube, cut the section rather than reaching in with a phone. A sawn window, photographed square to the cut, shows penetration, sugaring and any lack of fusion far more honestly than any endoscope image. Mark the cut line on the outside first so the window lands where the fault is.
Put a rule against the part in at least one frame. Not to measure the fault precisely, but to give the reader a sense of wall, gap and bead width. Without it, a hairline in a thin wall and a hairline in a heavy flange look identical, and the assessment starts with a wrong assumption about which part of the pipe failed.
Orientation matters just as much. Photograph the part the way it sits on the car, then a second time from above. Anyone reading the file is trying to match your picture to a drawing they hold, and the drawing has an up. A part photographed upside down and rotated will be misinterpreted, politely but firmly.
Keep the background plain. A bench covered in tools and swarf makes a small defect hard to find, and the reader spends attention on the clutter. A sheet of plain card behind the joint, or a clean patch of floor, is enough. Do not crop so tightly that the joint loses its neighbours.
Do not edit the images beyond rotation and cropping. No filters, no sharpening, no colour correction on a heat tinted part, because tint is evidence of how hot the metal got. A saturated photograph of a titanium weld can suggest an overheating problem that never existed, and that misleads the person trying to help you.
A complete set is small: the wide frame, two close frames of the fault, one frame with a rule, one of the part label, and one of the installation if it is still on the car. Name the files in order and keep them in one folder rather than scattered through a thread.
Then write a short note that fixes the facts the pictures cannot show: when it was fitted, what changed since, and what the car was doing when the fault appeared. The custom build record tells the reader what was intended; the photographs tell them what happened. Together they are enough. Anything less and the assessment stalls, which is why the fabricated parts range is documented the same way at dispatch.