How to photograph car damage for the most accurate AI assessment
The single biggest factor in how accurate your Chassly assessment is, more than the AI model, more than the vehicle data, is photo quality. The model can only assess what it can see clearly. A blurry photo, harsh shadow, or shot taken from a misleading angle can hide important detail or make minor damage look severe. Spending an extra two minutes to take good photos is the highest-leverage thing you can do.
The four shots every assessment needs
Chassly asks for 4-6 photos because cross-referencing multiple angles is how the AI builds confidence in its diagnosis. A single straight-on photo often hides depth, and a single close-up can't establish the location on the vehicle. The four core shots: a wide context shot, a close-up surface shot, and two angled shots from different sides of the damage.
Wide shot: stand back 6-8 feet (about 2 meters) and capture the entire damaged panel plus the panels next to it. This lets the AI identify which part is damaged ('front-right fender, not rear-right fender') by seeing it in context. Don't crop too tight; include enough surrounding sheet metal to anchor the location.
Close-up: get within 1-2 feet of the damage. Fill the frame with the actual damaged area. This is where the AI can see paint disruption, surface texture changes, and small details that determine repair vs. replace. A close-up shot taken at arm's length is usually about right.
Angled shots: take two photos of the same damage from different angles: one from the left side, one from the right side, or one from above and one from below. Angled photos reveal depth that a straight-on shot hides. A dent that looks 'just paint' from straight-on might be 2 inches deep when seen from a 45-degree angle.
Lighting: the easiest thing to get wrong
Direct overhead sun creates the worst conditions for damage assessment. Glare washes out paint detail, and harsh shadows can either hide real damage or create fake 'damage' (shadow lines that look like scratches). If you're outside on a bright day, move the car into shade or wait for golden-hour or overcast conditions.
Indoor garage lighting is usually fine, but watch for fluorescent-tube reflections that look like cracks. If your photos show streak patterns that aren't on the actual car, reposition slightly to avoid the reflection.
Phone flash is a mixed bag. It eliminates shadows but can cause harsh hot spots on glossy paint. If you must use flash, take both flash and no-flash versions and submit the clearer one. Modern phones (iPhone 13+ or any Pixel from the last few years) handle low light well enough that flash usually isn't needed.
Distance, focus, and resolution
Modern smartphone cameras have plenty of resolution for accurate AI analysis. Chassly automatically compresses photos before sending them to the AI model (typically to about 1280px on the long edge), so giant 12MP raw files don't help; they just upload slower.
What matters more than resolution is focus. Tap the screen on the actual damaged area before taking each shot. Phones default to focusing on the closest object in the center of the frame; if you're holding the phone at an angle, focus might land on the hood instead of the door damage you're trying to capture.
If your phone supports HDR, leave it on. HDR helps balance exposure across bright sky and shaded car body, which makes damage easier to see in mixed-light conditions.
Multiple damage areas: when to split assessments
Chassly supports up to 3 damage areas in a single assessment. If your car has separate damage in two or three different spots (say, a rear bumper scrape and a driver-door dent from unrelated incidents), you should add each as its own labeled area in the upload step.
Why split rather than just submit a pile of photos? Because the AI treats each area independently. If you mix photos from two areas in one batch, the model might assume they're related and try to draw connections that don't exist. Labeled areas tell the model 'these 3 photos are area 1, these 3 are area 2, analyze each independently.'
Each area needs at least 2 photos. Single-photo areas don't give the AI enough angles to triangulate depth confidently. If you only have one good photo of a particular area, get a second from a different angle before submitting.
Common mistakes that hurt accuracy
Taking all your photos from the same angle. The model has no extra information to work with; it's analyzing the same view repeatedly. Two well-chosen angles beat six identical ones.
Photographing wet damage. Water droplets hide surface texture and look like additional damage. Wait for the car to dry, or wipe the area with a microfiber cloth before shooting.
Including yourself or other people in reflections. The AI ignores reflections most of the time, but heavily reflective shots can confuse the part-identification step. Stand to the side or use a tripod.
Cropping too tight. If the AI can't see the panel boundary (fender meets door, door meets quarter panel), it can't tell you which specific part is damaged. Always leave some surrounding context in at least one shot.
Submitting heavily filtered or edited photos. Chassly is looking at original surface conditions. Filters that brighten paint or sharpen edges create artifacts that look like damage to the AI.
Frequently asked questions
Can I just submit one really good photo?
No. Chassly requires a minimum of 4 photos because single-angle photography systematically misses damage depth and hides parts of the affected area. The 4-photo minimum is a quality floor, not a soft suggestion.
What if the damage is on a curved surface like a fender peak?
Curved surfaces benefit even more from multiple angles. Take one photo straight-on, then move to a 45-degree angle on each side. Curves throw shadows that vary dramatically with viewing angle, so multiple perspectives are how the AI separates real damage from light play.
Should I clean the car first?
A quick wash helps but isn't required. Dirt and dust don't usually fool the AI; the system prompt explicitly tells the model to ignore dirt, road debris, and water droplets. But a clean panel makes paint disruption (scratches, scuffs) much more visible.
How close is too close for a close-up?
If you can't focus, you're too close. Most modern phones have a minimum focus distance of about 4 inches (10cm). Closer than that and the camera can't lock focus, producing a blurry image that the pre-screen will reject.
Can I use a friend's phone if mine has a broken lens?
Absolutely. The AI doesn't care which phone took the photos. Just make sure the lens is clean; a smudged lens softens detail across the entire image and reduces accuracy more than any other single factor.