Retrovision Labs
Retrovision Labs is a free browser-based tool that turns a color image plus a separately produced depth map into a navigable 3D mesh, exportable as GLB. Built for game artists, animators and designers who need fast 2D-to-3D previsualization.
What is Retrovision Labs?
Retrovision Labs is a small independent workshop that presents itself as an epic quest to reimagine the most memorable videogame classics with a new generation of creative tools. Its only public product so far is a browser application called IMAGE TO 3D, badged Preview v0.3.1 inside the interface and v0.3 on the home page button. It opens directly, with no account, no sign-up and no payment.
The principle is a depth-driven extrusion. You bring two images: a color map, the visible picture, and a depth map, the per-pixel distance information. The application reads both and builds a 3D mesh you can move through in real time. One point deserves emphasis: the tool does not estimate depth itself. Its Interpretation setting offers a single profile, "Default (MiDaS, Zoe, Leia, etc)", which names the third-party depth-estimation models whose output is expected as input. Those models are neither features of the product nor integrations with it, and several third-party directories get this wrong.
Accepted formats are .png, .jpg, .jpeg, .avif and .webp for the color map, and .png only for the depth map, decoded at 8 or 16 bits with ICC profile detection. Once both are loaded, the scene is shaped live through nine projection controls: Field of View, Distance Min, Distance Max, Power Near, Power Far, Skew X, Skew Y, Spherize and Tilt. Mesh density is governed by Triangulation, offering a "2 per depth pixel" option, and the result can be inspected in Wireframe. Two camera modes, Free and Focus, come with adjustable navigation speed and mouse sensitivity. The output is a GLB file, model/gltf-binary, generated in the browser and named after the two source files.
Technically it is a fully client-side PlayCanvas 1.70.3 application rendering through WebGL2 or WebGL1 behind a PCUI interface, and it warns that a depth map beyond 1-2 megapixels degrades interface performance. The marketing site is a single screen: logo, tagline, one button and three links, to Discord, to X and to an email address. Maturity is early. The domain was first captured in February 2024, the community Discord counted 38 members in August 2026, and the product is still labeled Preview.
What it does
- Turn a color image and a user-supplied depth map into a real-time 3D mesh
- Export the resulting mesh as a GLB file generated in the browser
- Shape the projection through nine controls: Field of View, Distance Min and Max, Power Near and Far, Skew X and Y, Spherize, Tilt
- Control mesh density with Triangulation and inspect the geometry in Wireframe
- Move through the scene in Free (fly) or Focus (orbit) camera mode, with adjustable Nav Speed and Mouse Sens
- Load 8-bit or 16-bit PNG depth maps alongside PNG, JPG, JPEG, AVIF or WebP color maps
- Work entirely inside the browser, with no account, no installation and no upload
When to use Retrovision Labs / When not to
A quick filter to help you decide if Retrovision Labs is the right fit.
When to use Retrovision Labs
- Game artists and developers who already hold a depth map and want to turn a flat 2D plate into a navigable 3D set
- Concept artists and art directors doing fast blocking and previsualization rather than final production assets
- 3D and motion artists comfortable with third-party depth-estimation models such as MiDaS, Zoe or Leia
- Freelancers and small studios who need a GLB file they can drop straight into a game engine or a DCC package
- Curious creatives looking for a free, account-free experiment with no commitment and nothing to install
When not to use Retrovision Labs
- Anyone expecting the depth map to be generated for them: it must be produced elsewhere and brought in
- Teams needing a watertight, full-volume reconstruction, since this is a single-viewpoint extrusion and the back of the scene stays undefined
- Developers looking for an API, a mobile app or third-party integrations, none of which exist
- Organizations bound by a compliance review, given the total absence of terms, privacy policy and identified publisher
- Users working at high resolution, as the app itself warns that depth maps above 1-2 megapixels degrade interface performance
How to use Retrovision Labs
A typical end-to-end flow, from setup to results.
- Open the web application from the single button on the home page: there is no account to create and nothing to install
- Prepare the depth map beforehand with a third-party depth-estimation tool, since the app expects one rather than producing it
- Follow the opening hint, "Start by uploading your custom maps below"
- Click Color Map and load the visible image (.png, .jpg, .jpeg, .avif or .webp)
- Click Depth Map and load the depth image (.png only, 8 or 16 bits), keeping the Interpretation preset "Default (MiDaS, Zoe, Leia, etc)" unless you have reason to change it
- Watch the mesh appear in the viewport, then work through the sidebar sections: Image, Color, Depth, Model and Preview
- Tune perceived depth with Distance Min, Distance Max, Power Near and Power Far
- Adjust the projection geometry with Field of View, Skew X, Skew Y, Spherize and Tilt
- Check mesh density with Triangulation and the Wireframe display, then navigate using View Mode Free for free flight or Focus for orbiting, with Nav Speed and Mouse Sens tuning the movement
- Click Export GLB to download the mesh, which takes its name from the two source files
Pros & Cons
Pros
- Free and immediate: no account, no sign-up, no payment, nothing to install
- Everything runs in the browser, and in the audited 0.3.1 build the images never leave the machine
- Standard GLB output, directly usable in a game engine or a DCC package
- Fine-grained projection control through nine parameters with real-time visual feedback
- Accepts depth maps from the models in common use, such as MiDaS, Zoe and Leia
- Works on any WebGL2-capable browser, with no operating system constraint
- An active community channel reachable directly from the application through Discord
Cons
- Does not produce the depth map, so an upstream third-party tool is always required
- Single-viewpoint extrusion rather than a complete volumetric reconstruction
- No terms of service, no privacy policy and no legal notice anywhere on the domain
- Unidentified publisher: no company name, no postal address, no named beneficial owner
- No written documentation, no tutorial and no features page: everything is learned inside the interface
- Practical resolution ceiling, the app warning that depth maps above 1-2 megapixels degrade performance
- No API, no mobile app and no integrations, with a single triangulation option and no mesh decimation setting
Pricing & Plans
The tool is free of charge in its entirety and offers a single, permanent level of access. No pricing page is published, the usual path returning 404 on both hosts, and the application contains no account, credit or payment mechanism, with no trace of any payment provider. There is consequently no paid tier, no lowest price point, no currency and no billing unit to report. No time-limited trial is mentioned either, the notion having no object where the whole product is already available at no cost.
Data, GDPR & hosting
A consolidated view of how Retrovision Labs handles your data.
GDPR overview
There is no mention of the GDPR anywhere on the domain. No privacy policy, terms of service or legal notice is published, and the usual paths for them return 404 on both hosts. No data controller is identified, no company name or address appears, no Article 27 EU representative and no data protection officer is named, no data processing agreement is offered and no sub-processor list exists. The only contact channel is a generic mailbox. Separately, and strictly as a technical observation rather than a publisher statement, the audited 0.3.1 build performs no server upload and has no account system, so the personal-data processing surface appears very small. That observation is not documentation: for GDPR purposes, nothing at all is formally set out by the publisher.
Who owns the data?
No ownership clause is published. The domain carries no terms of service, no privacy policy and no legal notice, and the usual paths for them return 404 on both hosts, so nothing states who owns the images you bring or the meshes you export, and the publisher asserts no claim either way. Separately, the audited 0.3.1 build performs no application network call: files are read locally through FileReader and URL.createObjectURL, and the GLB is assembled client-side and delivered as a Blob. Your files therefore never leave the browser and no server-side copy exists to be claimed. That is an observation about the audited build, not a contractual commitment.
Reuse rights
Nothing is declared. With no terms of service and no privacy policy, the publisher grants no license, imposes no restriction and offers no assurance regarding the reuse of what you import or export, so no permission can be said to have been given or withheld. Separately, the audited 0.3.1 build shows no server-side upload, no analytics or advertising tag, no cookie banner and no account system, which means there is no channel through which user content could reach the publisher and therefore nothing on its side to reuse. In practice your rights over the source images and over the exported GLB rest on whatever rights you already hold in the material you supply, not on any grant from Retrovision Labs. Again, this reflects the version audited on 23 August 2026 rather than a published position.
Data retention & training
Hosting summary
No hosting country or region is declared by the publisher. With no privacy policy, no terms of service and no legal notice on the domain, there is no statement about where anything is stored, and no jurisdiction is claimed. What can be observed concerns the infrastructure only: the domain resolves to 143.244.187.252, an address located in Santa Clara, United States, on AS14061 DigitalOcean LLC, without anycast, while the registrar is HOSTINGER operations, UAB, in Lithuania. The essential point is that this infrastructure serves static files. Analysis of the audited 0.3.1 application bundle shows no upload of user images: the color map, the depth map and the exported GLB are all handled inside the browser. There is therefore application hosting to describe, but no user-data hosting, since no jurisdiction can attach to content that never reaches a server. This is a technical observation about the version audited on 23 August 2026, not a commitment by the publisher, who documents nothing on the subject.
Things to keep in mind
Risks and trade-offs to weigh before adopting Retrovision Labs.
- No privacy policy, no terms of service and no legal notice anywhere on the domain: you use the tool without published rules, guarantees or recourse
- The publisher is unidentified, with no company name, no address and no named beneficial owner, so no one is formally accountable if anything goes wrong
- No GDPR mention, no data processing agreement, no sub-processor list and no declared minimum age, which makes the tool unsuitable wherever a compliance review applies
- Preview 0.3 status means no service commitment and no published roadmap, and the domain registration, held under a redacted registrant, runs only to February 2027: plan for the possibility that the tool simply disappears
- Nothing is saved on the publisher's side, so closing the tab without exporting destroys the work, which punishes the ordinary habit of leaving a session open for later
- The finding that everything runs locally comes from auditing build 0.3.1 in August 2026, not from any publisher commitment; it could change in a later version, so re-check before trusting the tool with sensitive or confidential imagery
- Results arrive fast and look convincing from the right angle, which can flatter: leaning on an extrusion instead of real geometry quietly erodes the habit of building and judging genuine volume
Setup & Integrations
Technical difficulty
Getting started is trivial: open a web page, with no account, no installation and no configuration. The real difficulty sits upstream, since you must already have a depth map, which means mastering a third-party depth-estimation tool before you arrive. The nine projection controls also assume some familiarity with field of view, near and far planes and projection in general, and there is no written documentation: you learn from the first-run hints and by trial. Hardware requirements are modest but real, namely a WebGL2-capable browser and a machine able to hold a scene built from a 1-2 megapixel image.
Deployment
Supported languages
Behind Retrovision Labs
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
What does Retrovision Labs actually do?
Does the tool generate the depth map for me?
Which file formats are accepted?
In which format does it export?
Do I need to create an account?
How much does it cost?
Are my images sent to a server?
Is there an API or a mobile app?
What depth map resolution should I use?
What state is the product in, and where can I ask a question?
Should you pick Retrovision Labs?
Retrovision Labs' IMAGE TO 3D is a narrow, free and genuinely useful tool, provided you understand what it is: one link in a 2D-to-3D chain rather than a complete solution. It earns its place when a depth map already exists, produced by MiDaS, Zoe, Leia or any comparable model, and you want to see that flat plate as a navigable set, tune the projection by eye and walk away with a GLB. Without a depth map you need another tool first, and that prerequisite, not the interface, is the real barrier to entry.
Expect an extrusion seen from a single viewpoint rather than a volumetric reconstruction: the back of the scene stays undefined, and the app itself advises staying under 1-2 megapixels. That makes it an instrument for previsualization and blocking, not a production pipeline.
The main reservation is not technical. Nothing legal is published on the domain, neither terms nor privacy policy nor legal notice, and the publisher is not identified: no company name, no address, no beneficial owner. For an individual experimenting with a free tool that matters little; for an organization with a compliance review it is disqualifying on its own.
Maturity matches: Preview v0.3.1, no published roadmap, no service commitment, a 38-member Discord and a generic mailbox as the only support. Treat it as an experimental workshop rather than an established service.
The upside is that trying it costs nothing and commits to nothing. No account, no payment, and in the build audited on 23 August 2026 no application network call at all, meaning the images stay in the browser. That is an observation about this version, not a promise from the publisher. On those terms it is well worth an hour of anyone's curiosity.
- Choosing a selection results in a full page refresh.
- Opens in a new window.