How to Convert a Photo into a 3D Model in 2026: Web App, API, or Self-Hosted TripoSR

The first thing that separates the three routes is how much stands between you and your first model. The web app wins outright here: with Meshy you open a browser, upload a photo, and get a model, with no install, no GPU, and no code. For a one-off conversion, or for anyone who is not a developer, that is the shortest path there is.

The Tripo API sits in the middle: there is no software to install, but you create an API key in the Tripo console and write code to send the image and poll for the result, so it assumes you can call a REST endpoint. Self-hosting TripoSR is the most involved to begin: you set up a Python environment, install the model and its dependencies, and ideally have a CUDA GPU, though it falls back to CPU. That setup is a one-time cost and buys you a free, private, unlimited pipeline, but it is a real cost, and if you only need one model, the web app is the honest recommendation.

Cost is where self-hosting pulls ahead, especially at volume. Because you run the open TripoSR model on your own hardware, converting a photo to a model costs nothing per generation beyond electricity, and there is no cap on how many you make. If you are converting hundreds or thousands of images, that free, unlimited ceiling is decisive, and it is the reason people self-host at all.

The hosted routes bill for convenience. The Tripo API is transparent and pay-as-you-go: credits cost $0.01 each ($1 buys 100), an image-to-model task runs about 20 to 50 credits, so figure roughly $0.20 to $0.50 per image, and Tripo currently offers about 300 free trial credits to start. Meshy is subscription-shaped rather than per-model: a free tier gives you a monthly credit and download allowance, and paid plans lift the caps, so its cost depends on your monthly volume rather than a per-image price. For occasional use both hosted routes are cheap; at scale, self-hosting is the one that does not grow with your usage.

If the photo you are converting is sensitive, a product prototype, a client asset, personal, where it goes matters. Self-hosting is the only route where the image never leaves your machine: TripoSR runs entirely offline, so nothing is uploaded, and you control the whole pipeline, versions, parameters, and output, with no third-party terms in the loop. For confidential work or air-gapped environments, that is not a nice-to-have, it is the requirement.

The two hosted routes are cloud services, which is exactly what makes them easy but also what you are trading. Meshy processes your upload on its servers under its account and terms, and the Tripo API sends your image to Tripo's endpoint under your API key. Both are normal, reputable SaaS arrangements, and for most public or low-sensitivity images that is completely fine. But if privacy or full control is the priority, self-hosted TripoSR is the route that keeps everything local.

When converting a photo is one step inside a larger product, a marketplace that turns listing photos into 3D previews, a pipeline that batches a catalog, the API route is built for it. The Tripo API is designed around automation: you authenticate with a key, submit an image as an asynchronous task, receive a task id, and poll for the finished model, and it exposes related operations like texturing, rigging, and format conversion as endpoints. Rate limits are managed by concurrency, so you scale by running tasks in parallel.

The other two are not automation-first. Meshy is a web application you drive by hand, which is ideal for one model but not for programmatically converting a thousand. Self-hosted TripoSR can absolutely be automated, but you build that yourself: wrap the model in your own service with a framework like FastAPI or Flask. That is powerful and free, but it is your code to write and run, whereas the Tripo API is a queue that already exists. Note the honest limit on the API side: completion is by polling, not webhooks, and there is no single batch call, you run concurrent tasks.

How finished the model looks the moment you get it back differs by route, mostly on textures. Meshy returns a textured model with PBR material maps from a single photo in about a minute, so for a newcomer it produces the most presentable result with no extra steps. The Tripo API also returns textured output and offers a PBR texture option, so its ceiling is similar, with the detail you choose per request.

Self-hosted TripoSR is the honest trade here: it reconstructs clean geometry fast, but it outputs vertex colors rather than PBR material maps, and there is no separate texture-synthesis pass, so straight out of the model it looks less finished than the hosted routes. For many uses, a base mesh you will retexture or 3D print, that is fine, and Stability even points users to its SF3D successor for game-ready output. But if you want a photorealistic textured asset with no post-work, the hosted routes lead, and the newer diffusion-based models generally produce higher-fidelity geometry than TripoSR's single-view reconstruction.

What you can do with the model next depends on the file you get. The Tripo API is the most flexible: it returns GLB natively and its conversion endpoint exports GLTF, OBJ, FBX, USDZ, STL, and 3MF, so one generation can feed a game engine (FBX), AR (USDZ), or a 3D printer (STL), with the honest caveat that STL drops textures and some formats are not available for rigged models. Meshy is close behind, exporting STL, 3MF, OBJ, GLB, FBX, and USDZ from a single generation.

Self-hosted TripoSR is the narrowest on purpose: it outputs a GLB or OBJ mesh, and you convert to other formats yourself in Blender or a converter. For most uses that is enough, GLB and OBJ open everywhere, but if you need a specific format like USDZ for AR or FBX for a game pipeline without a manual step, the hosted routes hand it to you directly. If you are converting a photo specifically to 3D print it, see our companion guide on getting a printable STL.

If you plan to sell or ship what you make, the license is as important as the mesh. Self-hosted TripoSR is the cleanest: plain MIT, with no user cap, no regional exclusion, and no attribution, so you can use and sell your output freely, and self-hosting means no per-item cost. That is the strongest position for a commercial pipeline. The one honest caveat is that the hosted convenience at triposr.org is a paid product, not free; the free, unrestricted route is running the open model yourself.

The hosted routes carry conditions. Meshy is free to start, but free-plan output is licensed CC BY 4.0, which permits commercial use only if you credit Meshy; on a paid plan you own the assets outright. The Tripo API is a paid service, so its output comes under Tripo's hosted commercial terms and your purchased credits rather than an open license. All three are usable commercially for most projects, but for no-strings, self-owned output, the MIT-licensed self-host route is the clean answer.