Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Clear out junk files and repair common Windows errors3Fix the driver behind crashes, sound loss and screen glitchesiTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
MapAnything on Replicate turns one or more photographs, video frames, or geometric JSON inputs into metric 3D scene geometry. Vufinder’s hosted listing is the quickest way to try it: upload your inputs, choose whether you want a GLB mesh, point cloud, normals, or point-scale data, and run the model without configuring a GPU.
This guide explains what the deployment does, which files it accepts, how to run it from the web interface or an API, what the outputs mean, when local execution makes sense, and how the Apache-licensed checkpoint used by the listing differs from the project’s other weights.
What MapAnything on Replicate actually is
MapAnything is a metric 3D reconstruction model from the Facebook Research MapAnything project. “Metric” means the model predicts scene geometry in a scale-aware form rather than producing only a relative depth ordering. Its documented design is a single feed-forward model that supports more than 12 reconstruction tasks, including multi-image structure-from-motion, multi-view stereo, monocular metric depth estimation, registration, and depth completion.
Free tools Windows power users keep installed
One-click scans. No signup required.
That capability list describes the range of tasks the model is designed to handle; it is not a ranking, accuracy benchmark, or promise that every task performs equally well in every scene.
#1 Best Overall
- 【Industrial-Grade Accuracy】Achieve single-frame accuracy up to 0.03 mm and volumetric accuracy of 0.03 mm + 0.05 mm x L(m), faithfully reproducing the finest surface details and complex geometries with exceptional consistency. Full-Field Structured Light accuracy reaches 0.08 mm; VCSEL mode delivers 0.10 mm @ 300-500 mm and 0.20 mm @ 500-800 mm. Engineered to meet the demanding requirements of 3D printing, reverse engineering, and precision modeling applications.
- 【Ultra-Fast Scanning & Robust Frame Rate】Multi-line Laser mode delivers up to 105 fps with NVIDIA GPU acceleration. Full-Field Structured Light mode achieves up to 5,000,000 points/s. The high frame rate ensures a smooth, uninterrupted scanning experience, especially suited for rapidly capturing large objects and complex scenes, significantly boosting overall workflow efficiency.
- 【AI-Powered & Photo-Grade Retopology】 AI object segmentation (Windows only) identifies your target in one click, tracks it throughout the scan, and auto-filters background noise — delivering clean data and streamlining post-processing. The patented 3D Gaussian Splatting converts point cloud and RGB data into true-to-life 1:1 photorealistic models; import photos from your phone or camera to apply real textures, then export in splat format for gaming, animation, and VR.
- 【All-Weather Outdoor Scanning】Multi-line Laser mode operates reliably up to 50,000 lux; with an outdoor filter attached, scanning remains stable in lighting conditions of up to 100,000 lux; VCSEL mode operates reliably in up to 100,000 lux ambient light. Designed to overcome lighting challenges, it provides consistent all-weather performance for construction sites, archaeological digs, and industrial fieldwork.
- 【5 Scanning Modes】NIR band supports Full-Field HD Scanning (markerless, fine structured light), Hybrid HD Scanning (dual projectors, fast high-quality modeling), and VCSEL Rapid Scanning (high-density pattern, fast markerless capture). Blue light band supports 30-Cross Laser Lines Scanning (handles high-reflectivity metals & dark objects) and Single-Line Deep Hole Scanning (captures from deep holes & narrow grooves). Five modes for comprehensive indoor and outdoor coverage.
Keep the project, deployment, and checkpoint separate
- MapAnything project: the research code and model family documented by the official repository.
- Vufinder deployment: a hosted version exposed through Replicate’s web interface and API.
- Pretrained weights: the checkpoint loaded by that deployment. The Vufinder model README identifies it as
map-anything-apache.
These layers matter when you evaluate licensing, setup, updates, or results. Using the hosted listing is not the same as installing the research repository, and a project’s code license does not automatically apply to every checkpoint.
What you can upload
The listing is a multi-input reconstruction workflow, not an image generator. You provide observations of a scene and the model estimates 3D geometry from them. A normal phone or existing camera can supply the images; the interface does not require a specialized camera.
| Input | Accepted formats | How the listing handles it |
|---|---|---|
| Images | JPG, JPEG, PNG, WEBP | Multiple images can be submitted for a reconstruction. |
| Video | MP4, AVI, MOV | The service samples frames according to sampling_rate; every nth frame is selected, while the first and last frames are always included. |
| Geometric input | JSON | Geometric information can be supplied alongside the visual inputs when it matches the interface’s expected schema. |
Uploaded images and sampled video frames are padded to a common aspect ratio and resized so their maximum dimension is 518 pixels. This preprocessing is performed by the listing, so very large source files do not pass through at their original dimensions.
Recommended Free Tools
Choosing images
Use views that show the same scene with meaningful overlap. A sequence that moves around an object or room gives the model more correspondence information than unrelated photographs. Avoid mixing images that depict different arrangements of furniture or objects unless that change is intentional for your task.
Choosing video
Video is convenient when you can walk a camera around a subject, but the sampling setting changes both the number of frames and the amount of visual redundancy. A lower numeric interval samples more frequently; a higher interval skips more frames. Because the first and last frames are included regardless of the interval, inspect those endpoints for blur, darkness, or an unintended view before running a long clip.
When geometric JSON is useful
JSON input is for workflows that already have geometric information to provide to the model. The listing accepts JSON, but the exact fields and structure are defined by its current input schema. Follow the live model form or API documentation rather than inventing keys from a different MapAnything implementation.
Rank #2
- EASY TO USE FOR BEGINNERS – Perfect for entry-level users, DIY creators, and 3D printing enthusiasts. Quick start, with simple practice giving optimal scan results.
- SMOOTH WIRELESS SCANNING – WiFi6-powered Ferret Pro ensures fast, stable scanning. Works with Windows, macOS, Android, and iOS for flexible cross-platform use.
- HIGH-PRECISION 3D MODELS – Capture detailed 3D models with full-color 24-bit scanning and anti-shake technology. Offers up to 0.1mm accuracy with full-color scanning. Ideal for objects from 50mm to 2000mm. Not suitable for very small or highly detailed items like jewelry or precision parts.cccc
- VERSATILE OUTPUT & ENVIRONMENT – Export in OBJ, STL, or PLY. Works reliably in most settings, including outdoor light (<30,000 lux). Avoid reflective, transparent, or very dark surfaces for best results.
- LIGHTWEIGHT & PORTABLE – Weighing just 105g, carry and scan anywhere—home, studio, or on the go. Compact, convenient, and ready for travel.
Run MapAnything from the hosted Replicate page
The hosted route is the best starting point if you want to understand the model before setting up Python, CUDA, or a local container.
- Open the current Vufinder MapAnything listing on Replicate. The live page is authoritative for the available inputs, model version, and output controls; version identifiers can change over time.
- Add your source data. Upload one or more supported images, a supported video, or the geometric JSON required by your workflow. For a video, set
sampling_rateto control how often frames are selected. - Choose the artifacts to return. Enable
return_meshwhen you need a surface model, and enablereturn_pcdfor a point-cloud result. The interface also exposes optional normal and point-scale outputs. - Start the prediction. The service preprocesses the inputs, runs the selected checkpoint, and reports the prediction status in the page.
- Download the results. The listing presents JSON output metadata and downloadable GLB mesh and point-cloud files when those outputs are enabled.
A GLB file is a practical interchange format for opening the reconstructed mesh in a compatible 3D viewer or modeling application. A point cloud is preferable when you want discrete 3D samples rather than a connected surface. Normals and point scales are supplemental outputs for pipelines that need orientation or per-point scale information.
How to select outputs for a first experiment
Start with a mesh when you need a visible surface
Enable return_mesh if your immediate goal is to inspect walls, furniture, an object, or another continuous surface in a 3D viewer. The resulting GLB is easier to navigate than raw numerical output for a first pass.
Use a point cloud for measurement-oriented inspection
Enable return_pcd when your downstream tool or analysis expects points. Point clouds preserve sampled geometry without requiring the model to connect every region into polygons.
Add normals or point scales only when your pipeline uses them
Normals describe local surface orientation, while point scales provide additional per-point scale information. They increase the information returned by the run but do not replace the geometric reconstruction itself. If you are only checking whether the scene shape is plausible, begin with the mesh or point cloud and add these fields after you know you need them.
Using the Replicate API
Replicate’s listing provides Python, Node.js, and HTTP examples. API calls require a Replicate API token and a model version identifier. Do not hard-code an identifier copied from an old tutorial as if it were permanent: version strings can change, so obtain the current identifier from the live listing when you deploy.
Rank #3
- 【High Accuracy & Fast Scanning】The Creality CR-Scan Ferret Pro 3D scanner delivers up to 0.1mm accuracy, 0.16mm resolution, and 30FPS scanning speed. It captures detailed dimensional data and complex shapes smoothly, creating highly realistic 3D models with ease.
- 【WiFi6 Wireless Transmission】Featuring advanced WiFi6, this handheld 3D scanner offers speeds 3x faster than WiFi5. The high bandwidth ensures stable, efficient data transfer for high-precision scanning and smoother workflow.
- 【Outdoor Scanning & Flexible File Export】Powered by upgraded optical technology and intelligent algorithms, the scanner delivers reliable performance in outdoor environments with ambient light up to 30,000 lux. Export models in OBJ, STL, or PLY formats for seamless integration with 3D printing, design, and reverse engineering workflows. For optimal results, avoid scanning highly reflective, transparent, or extremely dark surfaces.
- 【Anti-Shake Tracking】Equipped with one-shot 3D imaging, the Ferret Pro improves tracking accuracy and scanning success rates. Even with hand movements or quick object shifts, it ensures smooth, error-free scanning—perfect for beginners.
- Ferret Series Performance requirements: Windows: i5-Gen8 CPU or later Windows 10/11 (64-bit), RAM: >8GB, Software: >V2.3.0 Mac OS: M1/M2/M3/M4 series, macOS 11.7.7+ or Intel i5-Gen8+, RAM: >8GB Android: OS: Android 10.0+, RAM: >8GB, Connectivity: Wi-Fi 6, App: V2.0.2 iOS: Model: iPhone 11+, iOS 15+, RAM: >4GB
API workflow
- Create or use a Replicate account with API access and store the token in a secret or environment variable rather than embedding it in source code.
- Read the current Vufinder model page and copy its active version identifier.
- Send the image, video, or JSON input using the parameter names shown in that version’s schema. Include
sampling_rateonly for video workflows. - Set
return_mesh,return_pcd, and any normal or point-scale options according to the files your application needs. - Poll or await the prediction until it completes, then save the returned JSON metadata and file URLs to durable storage.
The exact request body and endpoint are intentionally version-specific. The live Python, Node.js, or HTTP example is safer than copying a static snippet because it reflects any schema or version changes made after this article was published.
Protect tokens and generated files
- Keep the API token on a server or in a protected CI secret; never place it in browser code or a public repository.
- Treat prediction outputs as application data. Download them to storage you control if they must remain available after the hosted prediction expires.
- Record the model version, input filenames, sampling setting, and enabled outputs with each run so you can reproduce a result after the listing changes.
Hosted execution versus running MapAnything locally
Replicate removes most infrastructure work, while local execution gives you control over the environment and avoids a per-run hosted charge. They are different operational choices, not two accuracy modes. Replicate and the official repository do not publish an accuracy comparison between them.
| Consideration | Hosted Vufinder listing | Official repository |
|---|---|---|
| Setup effort | Upload inputs and choose outputs in the web UI, or call the API. | Create an environment, install the project, and configure compatible machine-learning dependencies. |
| Hardware | Replicate supplies the inference hardware for the prediction. | You provide the computer, GPU, drivers, and storage. |
| Software versions | The listing controls the deployed image and checkpoint version. | You control the environment, but must resolve compatibility yourself. |
| Cost model | Usage is billed per hosted run; the displayed estimate changes with inputs and platform conditions. | No Replicate per-run charge, but you pay for your own hardware or cloud machine. |
| Operational control | Fast to start, with platform-managed updates and limits. | More control over repeatability, batching, and data location, at the cost of maintenance. |
Official repository quick start
The project’s quick start creates a Conda environment with Python 3.12 and then installs the project with pip. The repository does not pin PyTorch or CUDA, so installation is not a single universal command: select versions compatible with your operating system, GPU driver, and hardware.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallOutdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware match- Install Conda and a supported GPU driver if you plan to use CUDA.
- Create a fresh environment with Python 3.12, as specified by the repository quick start.
- Install the MapAnything project with pip inside that environment.
- Select and install compatible PyTorch and CUDA builds; the project leaves those versions unpinned.
- Run the repository’s example or inference entry point and verify it with a small input before processing a larger collection.
Replicate also documents local Docker/Cog use. That route packages an inference service for local execution but is more involved than trying the hosted deployment. It is useful when you need a repeatable containerized service and are prepared to manage the image, GPU runtime, and local storage.
Cost and run-time expectations
The Replicate page displayed an estimate of approximately $0.018 per run and a typical completion time of about 19 seconds on L40S hardware on October 13, 2026. These are platform estimates, not fixed prices or service-level guarantees. Replicate states that both price and prediction time can vary significantly with the inputs.
| Displayed estimate | Qualification |
|---|---|
| Approximately $0.018 per run | Replicate estimate checked 2026-10-13; varies with inputs and platform conditions. |
| Approximately 19 seconds | Typical completion estimate on L40S hardware; actual duration varies with inputs. |
For budgeting, multiply the current per-run estimate by the number of predictions you expect, then allow for retries and exploratory runs. Check the live listing immediately before committing to a production budget because hosted prices and hardware estimates are volatile.
Rank #4
- AI Visual Tracking Technology: Equipped with a new generation of single-frame encoded structured light units that improve surface feature detection for smooth, marker free scanning
- High Precision Scanning: Achieves 0.05mm accuracy using blue light technology, making your model size closely approach reality for detailed reproduction
- Fine Resolution Capability: Features 0.10mm resolution with detailed point clouds that preserve object details providing refined models for printing or display projects
- Versatile Scan Range: Capable of scanning objects ranging from 15mm to 1500mm, accommodating a wide variety of object sizes
- Integrated Software Solution: JMStudio scanning software integrates scanning, editing, and optimizing into one seamless process for efficient workflow
Licensing: code and weights are not the same thing
The MapAnything repository states that its code is licensed under Apache 2.0. The pretrained checkpoints have separate terms:
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
| Component | License identified in the project materials | Practical implication |
|---|---|---|
| Repository code | Apache 2.0 | The code license governs the source code, subject to its notice and attribution requirements. |
facebook/map-anything weights |
CC-BY-NC 4.0 | Non-commercial restrictions apply; review the full license before using these weights in a commercial workflow. |
facebook/map-anything-apache weights |
Apache 2.0 | The repository recommends this variant for commercial use, subject to Apache 2.0 terms and any other applicable obligations. |
| Vufinder Replicate deployment | README identifies map-anything-apache weights |
The deployed checkpoint is identified as the Apache-weight variant, but you should still review the current service and model terms for your use case. |
The official Hugging Face card identifies facebook/map-anything as the CC-BY-NC-4.0 variant. Do not infer the license of one checkpoint from the license of another, and do not treat a hosted service’s checkpoint notice as a substitute for reviewing the current terms that apply to your application.
Accuracy, scene quality, and sensible expectations
MapAnything’s documented task breadth makes it useful for experiments that combine different image and geometry sources, but the official project and listing documentation do not provide a verified comparative accuracy ranking or a task-by-task limitation table. It would be misleading to call this deployment the best 3D reconstruction model or to promise reliable metric accuracy for every scene.
Factors that can reduce reconstruction quality
- Insufficient overlap: if adjacent views share little visual content, the model has fewer correspondences for estimating camera relationships and geometry.
- Motion blur or poor exposure: smeared or very dark frames remove the visual features needed for reconstruction.
- Changing scenes: moving people, rearranged objects, or changing lighting can make a set of images internally inconsistent.
- Video sampling choices: skipping too many frames can remove useful intermediate viewpoints; sampling every frame can increase processing work without guaranteeing a better result.
- Input preprocessing: the service pads and resizes all submitted images and sampled frames to a 518-pixel maximum dimension, so fine details smaller than that preprocessing scale may not survive.
Validate a reconstruction visually and, when scale matters, against a known reference in your own workflow. The model’s capability description is not a substitute for application-specific testing.
Troubleshooting common first-run problems
The upload is rejected
Confirm that each file is JPG, JPEG, PNG, WEBP, MP4, AVI, MOV, or JSON as appropriate. Remove unsupported sidecar files and check that a video is not using a container outside the listed formats. If a JSON input fails, compare its fields with the current listing schema rather than assuming that arbitrary 3D data will be accepted.
The result is sparse, distorted, or incomplete
Try a shorter, sharper capture with more overlap between views. For video, adjust sampling_rate so the model receives enough intermediate frames. Check the first and last frames because they are always included and can introduce a bad viewpoint if they are blurred or unrelated to the main sequence.
Best Value
- HIGH ACCURACY & FASTER: Boasting an impressive accuracy of up to 0.1mm, a resolution of 0.16mm, and a scanning speed of 30FPS, the Creality CR-Scan Ferret SE 3D scanner demonstrates outstanding performance in capturing extensive dimensional data and intricate details to shape highly realistic models smoothly and quickly
- ANTI-SHAKE TACKING: CR-Scan Ferret SE 3D scanner equipped with the new one-shot 3D imaging technology, this advanced feature enhances tracking efficiency and significantly increases the success rate of scans , ensures smooth, error-free scanning results even with shaky hands or rapid movements, ideal for beginners
- COLORFUL & VIVID TEXTURES: The CR Scan Ferret SE color 3D scanner built-in with the 2MP high-resolution color camera, captures intricate details and colored 3D models in their original colors, vividly bringing every intricate detail to life
- FLEXIBLE SCANNING RANGE: Provides a flexible scanning range of 150mm to 2000mm and a single capture range of up to 560*820mm, easily and efficiently handles the scanning of medium to large objects
- SCAN BLACK/METAL OBJECTS WITHOUT SPRAYING: The Ferret SE is optimized for scanning black or metal objects, it doesn't required you to use a white powder or spray to create a contrasting surface for black objects, much easier and faster to help you to finish your work
The output file I need is missing
Return types are controlled by the interface. Enable return_mesh for a GLB mesh or return_pcd for a point cloud before starting the prediction. Normals and point scales are optional outputs and must be selected when your workflow requires them.
An API request fails authentication
Verify that the Replicate API token is present in the server-side environment and has not been exposed, expired, or copied with extra whitespace. Do not confuse the token with the model version identifier; both are required for a versioned API call.
A local install fails around PyTorch or CUDA
The repository intentionally leaves PyTorch and CUDA unpinned. Check your GPU driver, operating system, Python 3.12 environment, and the compatibility matrix for the PyTorch build you selected. A clean environment is usually easier to repair than repeatedly changing packages in an existing one. If local dependency work is not justified, use the hosted listing instead.
A practical beginner workflow
- Capture a small set of overlapping photos or a short walk-around video with steady movement and consistent lighting.
- Run the hosted listing with the default-style output you need first: mesh for visual inspection or point cloud for point-based processing.
- Inspect the GLB or point-cloud download in a compatible viewer and note missing surfaces, scale problems, or obvious artifacts.
- Improve the capture or video sampling before changing many model options; input quality is easier to diagnose than a large set of simultaneous setting changes.
- Record the model version, input set, sampling rate, and enabled outputs for any run you may need to reproduce.
- Move to API automation only after a manual run produces an output your application can consume.
- Choose local repository or Docker/Cog execution when you have a clear need for environment control, data locality, repeated batch processing, or a predictable deployment surface.
When MapAnything on Replicate is a good fit
Use the hosted deployment when you want a low-friction way to turn existing images or video into inspectable 3D geometry, test the model’s multi-input workflow, or prototype an application without managing a GPU. Its image, video, and geometric-input options also make it suitable for experiments that do not fit a single monocular-depth call.
Choose a local setup when you are prepared to manage Python, PyTorch, CUDA, containers, and model files in exchange for more control over execution. Whichever route you choose, identify the exact checkpoint and read its license: the code, the CC-BY-NC weight variant, the Apache weight variant, and the hosted service are related but not interchangeable.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
