Clients do not ask for a "3D product." They ask for something that looks like the site, or something they can measure, or both — and they use the same words for jobs that are not the same job.
That is how gaussian splat vs mesh fights start. Someone saw a photoreal orbit online. Someone else has a CAD template that expects a triangle surface. The scope says "3D model." You are left deciding whether to sell a walkthrough that looks right near the flight path, or a controlled mesh that survives a cut-fill argument.
If you still need the one-sentence definition of a splat, start with What Is a Gaussian Splat? (And What It Is Not). This post is the decision piece: when each deliverable makes sense, what fails when you swap them, and where a point cloud still belongs in the middle.
SkyView is Pilot Ledger's desktop processing lane for commercial pilots — local photogrammetry on your hardware, no per-map cloud processing fee, waitlist for September 2026. Gaussian splats stay on the desktop. Orthos and finished layers can share in a browser. Join the waitlist at pilotledger.com/#pricing. It is not available today.
Stop treating "3D" as one product
A mesh is a surface. Triangles (or similar faces) that approximate the ground, a stockpile, a facade, or a roof. When that mesh was flown and controlled for mapping, you can often measure on it, drape texture, export something a civil or BIM workflow recognizes, and defend numbers against check shots.
A gaussian splat is an appearance model. Soft 3D blobs optimized so a viewer looks photo-real from the camera poses you actually flew. It is built to look right when you orbit. It is not a triangle surface you treat as geometry by default. Looks right is not measurable.
A point cloud is discrete samples with coordinates — dense photogrammetry or lidar — often the language of classification, volumes, and "send me the LAS." It is not a splat. Do not confuse a photoreal orbit with points you would classify or volume from.
Orthos still sit beside all three. Plan view for overlays and sheets. Free-orbit 3D for walkthrough. Mesh or cloud for the products scopes still name when money or liability is on the line.
If your quote collapses those into one line item called "3D," you already lost the scope fight. Name the deliverable.
When a gaussian splat is the right deliverable
Sell a splat when the client is buying look and walkthrough, on purpose.
- The superintendent wants to spin the site in a meeting without flying out.
- The owner wants a photoreal orbit for stakeholders who will never open CAD.
- An indoor plant or building walkthrough matters more than stakeout vertices.
- The scope says walkthrough, visualization, or "looks like the site" — not survey-grade surface.
A good splat earns its fee near the path you flew. Overlap, multiple heights, and honest coverage still matter. Away from those viewpoints, the renderer invents. That is not a bug you paper over in the sales deck. Tell the client up front: near the path, photoreal; off the path, guesswork.
Do not sell a splat as a substitute for ortho, DEM, controlled mesh, or a volume report. Do not extract stakeout coordinates from a splat viewer. Do not claim absolute accuracy because foliage looks real. The earlier post already draws that line. This one is about picking the product before you fly.
When a mesh still wins
Sell a mesh when the client is buying surface they can argue with.
- Cut-fill, stockpile volume, grade check, facade quantities tied to a surface.
- Export into CAD, GIS, or BIM that expects a mesh or surface model.
- Independent check shots, a report, and a story about how the model sits on control.
- The scope names mesh, DSM/DTM, or "measurable 3D model" and the contract language matches.
A pretty texture on a mesh does not make it a splat. A photoreal splat does not make it a mesh. If the next person on the job needs a surface they can click and trust relative to control, you are in mesh (or point-cloud) territory.
Mesh quality still depends on flight design, control, and processing discipline — the same GNSS story as any mapping job (RTK, PPK, PPP and OPUS). A splat does not forgive a float-heavy flight any more than a mesh does. Appearance models and survey products both inherit bad poses.
Where point clouds fit (without stealing the H1)
Point clouds are not the keyword of this post, but pilots keep asking where they sit in a gaussian splat vs mesh decision.
Use a dense cloud when the workflow is classification, breaklines, volume from points, or a client who already lives in LAS/LAZ. Use a mesh when they need a continuous surface for CAD or sheets. Use a splat when they need the orbit that looks like a photo.
Same photos can feed more than one product. Same flight is not the same deliverable. If the client wants a walkthrough and a volume, price both. Do not hope one file answers both questions.
What fails when you swap them
Splat sold as mesh. The meeting looks great. Then someone asks for a quantity, a setback, or a stakeout point from the viewer. You either refuse and look like you oversold, or you hand out invented coordinates and own the liability.
Mesh sold as splat. The surface is honest. The client wanted a photoreal orbit for executives. The textured mesh looks "3D software," not "I am standing on the site." Wrong product for the buyer, even if the geometry is better.
One flight pattern for both, no plan. Mapping grids optimized for ortho and DSM are not automatically the capture you want for a photoreal splat. Indoor walkthrough capture is not automatically a survey mesh flight. If you promise both from one casual pass, say which product is primary and which is a bonus — in writing.
Vague scope language. "3D model," "digital twin," and "as-built model" are not deliverable specs. Rewrite them to splat walkthrough, controlled mesh, point cloud, ortho, or a named package. Your future self will thank you when the change order arrives.
How SkyView fits without pretending it is live
SkyView is the processing front door on Pilot Ledger: run photogrammetry on your own desktop, keep raw imagery on the machine, upload finished deliverables when you are ready to share. No per-map cloud processing fee. Waitlist for September 2026 — not available now.
What that means for this comparison:
- Meshes, orthos, and mapping products are the commercial mapping lane pilots already sell. Process locally, then share finished layers. Browser share for orthos is tiled viewing without forcing a giant whole-file download story on the client.
- Gaussian splats stay desktop-only in SkyView. Do not promise a web splat orbit as the client share. If the deliverable is a splat walkthrough, plan the handoff around desktop viewing — not a browser link you invent for marketing.
- Pilot Ledger quoting, jobs, and payment sit behind that. Second beat only. The H1 is still the deliverable decision, not CRM.
Join the September 2026 waitlist at https://pilotledger.com/#pricing. Ask follow-ups on Ask Mav on the same site when a format or scope line is still fuzzy.
A practical pick list before you quote
- What did the client name — walkthrough look, measurable surface, points, plan-view ortho, or a stack?
- Who opens the file next — superintendent phone, CAD tech, surveyor, or owner slide deck?
- Are check shots and control part of the money conversation?
- Are you pricing one product or two from related imagery?
- Can you say out loud, in one sentence, which product is primary?
If you cannot answer those without hedging, you are not ready to pick gaussian splat vs mesh. Fix the scope before you fix the processing settings.
Bottom line
A gaussian splat sells appearance near the path you flew. A mesh sells surface you can defend when the numbers matter. Point clouds still own a lot of classification and volume workflows. Orthos still own the sheet. Pick the deliverable the job actually bought.
SkyView is on the waitlist for September 2026 for pilots who want that processing on their own hardware without a per-map cloud fee — pilotledger.com/#pricing. For the definition of a splat itself, keep What Is a Gaussian Splat? (And What It Is Not) next to this post.