Fill 3D
Fill 3D is an AI virtual staging platform for real estate: upload a photo of an empty room, write a prompt, and get a photorealistic furnished image plus a downloadable Blender scene. Currently in beta.
What is Fill 3D?
Fill 3D is an AI-powered platform for virtual staging, designed to significantly speed up staging workflows for real estate professionals. You give it a photograph of an empty room and a sentence describing the furniture you want, and it returns a photorealistic image of that room furnished, plus, if you ask for it, a 3D scene you can open in Blender.
Two use cases are declared. 3D Reconstruction turns a single image of an empty room into a Blender scene with faithful geometry, lighting and materials. Virtual Staging takes an image plus a prompt, reconstructs the room in 3D, places 3D furniture models and renders the final picture.
The pipeline chains two models. blend-1 generates the 3D scene; stager-1-preview then chooses and positions furniture from the prompt and renders the image with Blender's photorealistic engine. That architecture is the product's stated position: rather than repainting pixels with diffusion inpainting, Fill 3D imitates the professional staging workflow. The documentation is explicit about why, noting that other AI virtual staging platforms generate disconcerting artifacts like warped geometry, altered colors, and more. Fill 3D is specifically designed to avoid these artifacts.
Published timings are 30 to 50 seconds for the 3D scene, and 2 to 3 minutes to stage and render a room. The reference point the documentation offers is a professional lifting an image into a 3D scene by hand, estimating camera parameters and modelling surfaces, in 30 to 60 minutes.
Input conditions are documented and strict: a clear floor, visible walls, broadly rectangular rooms, good lighting and a level camera, with a tripod recommended. The scope is narrow as well. stager-1-preview currently only stages with bedroom furniture, and a banner on the home page states that during the beta the service will generate staged images at less than full resolution. On the review date the app's model picker exposed a model named Fill 3D stager-1-nano, while the documentation covers only blend-1 and stager-1-preview.
Results can be shared through a public link that freezes the prompt history and the generations at the moment it is created; the documentation suggests using it to hand the Blender scene to an outside editor. The project was presented by its creator on Hacker News on 28 September 2023, and its open-source API starter carries an MIT licence and 743 stars.
What it does
- Virtually stage an empty room from a text prompt
- Reconstruct a room photograph into a usable Blender 3D scene
- Download the Blender scene to finish the staging by hand
- Share a public link to the conversation and its generations
- Call generation from your own code through the REST API
When to use Fill 3D / When not to
A quick filter to help you decide if Fill 3D is the right fit.
When to use Fill 3D
- Real estate agents and brokers who need to furnish an empty room for a listing photograph.
- Agencies and real estate photo editing studios that outsource staging: the generated Blender scene can be shared and retouched by hand.
- 3D artists and archviz professionals who want a geometrically correct Blender base without lifting the image into 3D manually.
- Developers embedding staging in their own application through the public REST API, with an open-source Next.js starter project to begin from.
- Teams that need to pass a result to a colleague, a client or an outside editor: a share link carries the prompt history and the generations.
When not to use Fill 3D
- Rooms that are already furnished: the floor must be clear, and the documentation states that support for item removal is still being worked on.
- Anything other than bedrooms for automatic staging: stager-1-preview currently places bedroom furniture only.
- Awkward geometry and poor capture conditions: the documentation says it works best with rectangular rooms, well-lit photographs and a level camera.
- Anyone who needs full-resolution deliverables: during the beta, staged images are generated at less than full resolution.
- Buyers who need a contractual or compliance commitment: no terms of service, no privacy policy and no data processing agreement are published.
How to use Fill 3D
A typical end-to-end flow, from setup to results.
- Go to fill3d.ai and create an account by simply logging in.
- Pick one of the sample images, or upload a photograph of an empty room (the input accepts JPEG: image/jpeg, .jpg, .jpeg).
- Select the model: blend-1 for 3D reconstruction, or the staging model.
- Write your prompt following the documented structure, in this order: object, then location, then orientation, then scale. Press enter to run it.
- Use the documentation's own phrasing as a template: Place a large bed with nightstands and an orange bed cover, against the center of the left wall, between the two windows, with the bed rotated so that it is facing the opposing wall, and make sure it is large enough to fill much of the floor space.
- Remember the placement rule: an object placed against a wall has its orientation locked flat against that wall.
- Repeat the sequence for each object you want to place.
- Download the generated Blender scene if you want to finish the staging by hand or hand it to an editor.
- Use the Share button to create a public fill3d.ai/share/ link; it freezes the messages that exist at that moment, so adding more means deleting and recreating the link.
- For the API route: generate an API key in the app (five active keys maximum), purchase API credits, then POST to https://www.fill3d.ai/api/v1/generations with the header Authorization: Bearer followed by your API key.
Pros & Cons
Pros
- A 3D and path tracing approach rather than diffusion inpainting, which keeps lighting and perspective consistent with the original photograph.
- Output you can work with downstream: the generated Blender scene is downloadable and editable by hand.
- Time savings quantified by the publisher's own documentation: 30 to 50 seconds against 30 to 60 minutes for manual 3D lifting.
- Structured prompting gives explicit control over the object, its position, its orientation and its scale.
- A documented public REST API with an open-source starter project under the MIT licence.
- Clear documentation that spells out the image conditions to respect before uploading anything.
- Share links let a colleague, a client or an outside editor pick up the conversation and the scene.
Cons
- Declared beta: staged images are generated at less than full resolution.
- Staging is limited to bedroom furniture with stager-1-preview.
- The input photograph must show an empty, well-lit, broadly rectangular room shot with a level camera.
- No removal of existing furniture: the documentation presents item removal as still being worked on.
- No live pricing page, so the cost of use cannot be checked before creating an account, and subscription billing is not active.
- No terms of service, no privacy policy, no legal notice, no company name and no postal address are published anywhere.
- No compliance material either (no GDPR mention, no DPA, no sub-processor list, no retention period), the whole product sits behind authentication, the Discord invite announces 34 members and the blog is still empty.
Pricing & Plans
No price is published. As of 16 August 2026 the /pricing page returns a 404, as do /terms, /privacy and /about. A pricing page did exist between September 2023 and February 2024 according to the Internet Archive, but its content could not be retrieved. The site's authentication backend (Clerk) answers that the billing feature is not enabled for this instance, so subscription billing is not active. The only live paid mechanism documented is API credits: the API reference states that you must purchase API credits to create generations using the API, and the balance is shown in the app's API Keys dialog. No amount, no currency and no credit tier is published for those credits. A plan table does still exist inside the JavaScript shipped by the site, describing a Starter plan at 49 monthly or 499 annually with 100 generations per month, commercial use and community support, and a Business plan at 99 monthly or 999 annually with 250 generations per month and multiple viewpoints, marked disabled behind a Coming soon button. That table is referenced nowhere in the code, is never displayed, and carries no currency symbol and no currency code. These figures are therefore reported as code found on the site, not as an offer: no price, no currency, no free plan and no free trial can be stated for this tool at the review date.
- there is no live pricing page
- and subscription billing is disabled on the site's authentication backend.
- API credits are the only live paid mechanism documented. The API reference states they must be purchased to create generations
- but no amount
- no currency and no credit tier is published.
- Starter
- 49 monthly or 499 annually (no currency appears anywhere in the code)
- 100 generations per month
- commercial use
- community support.
- Business
- 99 monthly or 999 annually (no currency appears anywhere in the code)
- 250 generations per month
- multiple viewpoints
- everything from Starter
- marked disabled behind a Coming soon button.
- nothing in the application references them
- and they carry neither a currency symbol nor a currency code. They must not be read as current prices.
Data, GDPR & hosting
A consolidated view of how Fill 3D handles your data.
GDPR overview
There is no mention of the GDPR anywhere on the site, in the documentation or in the app. No Article 27 representative is designated, no data protection officer is named, and no privacy contact address is published; the only address published anywhere is hi@fill3d.ai, presented as support. No data processing agreement is offered, no sub-processor list exists, and no retention period is stated. The publisher itself is not identified: no legal entity name and no postal address appear on any page, and the Clerk authentication backend returns privacy_policy_url = null and terms_url = null. For a user in the European Union, this means there is no published basis on which to exercise access, deletion or portability rights, and no documented counterparty to address a request to. This is a statement of what is absent as of 16 August 2026, not an allegation of non-compliance.
Who owns the data?
No ownership terms are published. Fill 3D has no terms of service and no privacy policy on its site or in its documentation, and the Clerk authentication backend it uses returns terms_url = null and privacy_policy_url = null. Nothing therefore establishes who owns the photographs you upload or the images and 3D scenes the models return, and nothing states what the publisher may do with them or with whom it may share them. The only contract-like wording found anywhere in the product is a Commercial use line inside a plan definition buried in the site's shipped JavaScript, which is neither displayed nor sold. Share links, finally, make the frozen messages and generations visible to anyone holding the URL.
Reuse rights
Nothing is documented. No text on the site, in the documentation or in the app describes what is done with the photographs you upload or with the prompts you write, and no reuse right is granted or withheld in writing. Targeted searches for wording about model training, opt-out, do not train or zero retention returned no occurrence in either direction: the publisher neither claims to train on customer data nor states that it does not. What can be observed is technical only. The app is a Next.js front end served from Vercel infrastructure (IP 76.76.21.21, AS16509 Amazon, United States), authentication runs on Clerk, images are served from cdn.fill3d.ai, and the creator stated on Hacker News in 2023 that compute runs on Function (fxn.ai). These are network-level observations, not declarations by the publisher. In practice a user has no written permission to rely on and should assume nothing about reuse rights until terms are published.
Data retention & training
Hosting summary
No hosting country and no hosting region is declared on the site or in the documentation. What follows is observed, not stated. On 16 August 2026, fill3d.ai resolved to 76.76.21.21, an anycast address assigned to AS16509 (Amazon.com) and geolocated in the United States, in Walnut, which is the signature of Vercel hosting. Images are served from cdn.fill3d.ai, the documentation is hosted by Mintlify, the blog runs on Substack, and authentication runs on Clerk. The creator stated on Hacker News in September 2023 that GPU execution runs on Function (fxn.ai). Each of these is a network-level or third-party observation, not a declaration by the publisher: none carries a commitment about where uploaded photographs or generated images are stored, and none can be relied on as a data localisation guarantee. For a European user in particular, no hosting jurisdiction, no transfer mechanism and no sub-processor list is published.
Things to keep in mind
Risks and trade-offs to weigh before adopting Fill 3D.
- No contractual document is published: no terms of service, no privacy policy, no legal notice. The Clerk authentication backend returns privacy_policy_url = null and terms_url = null, so there is nothing to read and nothing to agree to.
- No legal entity, no postal address and no country of establishment are published, so there is no identified counterparty to turn to if something goes wrong.
- Nothing is stated about what happens to the photographs you upload: retention, model training, sub-processors and hosting location are all undocumented.
- Prices cannot be consulted before creating an account, which makes it impossible to size a pilot in advance.
- Sensitive use: virtual staging produces visuals for property listings, and the Hacker News launch thread carries a long discussion of the risk of misrepresenting a property. Disclosing that an image is virtually staged remains the user's responsibility, not the tool's.
- Uploaded images are served from cdn.fill3d.ai, and share links are public to anyone holding the URL. Treat a shared link as published, not as private.
- Leaning on automatic furniture placement can dull the habit of checking scale and plausibility yourself. A render that looks right is not a measurement of the room.
Setup & Integrations
Technical difficulty
Low for the web route: you create an account by simply logging in, upload an image and type a prompt, with nothing to install. The real difficulty lies elsewhere, in writing a structured prompt (object, location, orientation, scale) and in supplying a photograph that meets the documented conditions. Working with the generated 3D scene assumes you know Blender. The API route targets developers: generate a key in the app, purchase API credits, then POST to /api/v1/generations with a Bearer header. An open-source Next.js starter project can be cloned to shorten that step.
Deployment
Integrations
Behind Fill 3D
Social
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
What does Fill 3D actually do?
How long does a generation take?
What kind of photograph should I provide?
Can it remove furniture that is already in the room?
What furniture can it place?
Is there an API?
Can I share a result?
How much does it cost?
Are there terms of service or a privacy policy?
Is the product finished?
Should you pick Fill 3D?
Fill 3D makes a clean, differentiated technical bet: reconstruct the room in 3D and render it with path tracing instead of repainting pixels with diffusion inpainting. On the evidence available, that bet holds together. Lighting and perspective stay consistent with the original photograph, and the output is not a dead end, since the Blender scene can be downloaded and finished by hand. The time savings the publisher documents are concrete: 30 to 50 seconds against 30 to 60 minutes of manual 3D lifting, and 2 to 3 minutes to stage and render a room.
The product is still narrow. It is a declared beta rendering at less than full resolution, it places bedroom furniture only, it needs a genuinely empty, well-lit, broadly rectangular room, and it cannot remove existing objects. Anyone whose listings are not empty bedrooms shot on a tripod should test before planning around it.
The harder reservation is not about the product. The publisher is opaque: no legal entity is named, no postal address is published, and there are no terms of service, no privacy policy and no legal notice, with the authentication backend itself returning null for both document URLs. No price is published either, so the cost of use cannot be assessed before creating an account. Signals of activity are mixed: the JavaScript served by the site was rebuilt in 2026 and the app exposes a stager-1-nano model, but the public GitHub repository has had no push since December 2024 and the blog is still empty.
Worth trying if you are a real estate professional, an archviz artist or a developer curious about a 3D-first approach to staging, and if you can accept handing over images and prompts without a written contract. Not yet a tool to build a client-facing pipeline on.
- Choosing a selection results in a full page refresh.
- Opens in a new window.