From Prompt or Photo Through Mesh Density to a PBR GLB

A usable 3D generation session is a short pipeline, not a single button. You choose an input, you choose how dense the mesh should be, and you leave with a textured asset in a format an engine can load. This note walks that path as it appears on the live pages of 3DMaker, without treating marketing numbers as measurements.

Input stage: a prompt or a photo, not both contracts on one screen

The marketing homepage is only the briefing. Its heading calls the product an AI 3D model maker for any idea, and a section titled One AI 3D Model Generator, Two Ways to Start says a written description and a single photo are both supposed to produce the same textured, game-ready mesh. The homepage gallery even invites you to drag to rotate sample meshes. The actual input contracts, though, live on two tool pages.

The Text to 3D page is the prompt stage. You get up to 1,024 characters for materials, proportions, and style. No sketch or reference photo is required. A Surprise me control can supply a starting prompt when you do not yet have one. The page’s guidance is to be specific: a weathered or dented object description gives the generator more to use than a bare noun.

The Image to 3D page is the photo stage. It leads with a single reference image and accepts PNG, JPG, or WEBP. The page says PBR materials are pulled from that source photo, and that a plain, well-lit background helps the subject separate from its surroundings. The homepage mentions that an image path can use up to four angles; this dedicated landing still frames the job as one photo, so I am not treating multi-angle capture as the default step in this pipeline.

Density stage: Fast, Standard, Pro, or Ultra

Both tool pages expose the same four mesh density names: Fast, Standard, Pro, and Ultra. They are UI options for how much detail the mesh should carry, not a timing table. The text page says you can switch density without changing the prompt. The image page says you can switch density without re-uploading the photo. In a pipeline, that matters more than the labels themselves: input and density are independent knobs. I am not citing the product’s own totals block, or any stated generation time, as independent data.

Output stage: PBR on the mesh, GLB plus a preview

The visible export contract is consistent across the homepage FAQ and both tool pages: the generator textures the mesh automatically and exports GLB and a rendered preview image. PBR textures are included; the pages do not describe a separate texturing pass. The destination list is Blender, Unity, Unreal Engine, Three.js, Babylon.js, or model-viewer. I am not treating other format names that appear only in social-preview text as an on-page export promise.

That output choice matches how the industry talks about runtime delivery. The Khronos Group describes glTF as a royalty-free specification for the efficient transmission and loading of 3D scenes and models by engines and applications. It minimizes asset size and the runtime work needed to unpack the file. The core description is JSON; a scene can also be stored as a single binary .glb. glTF 2.0 is published as ISO/IEC 12113:2022, and Khronos PBR is the material model used inside glTF to approximate real-world lighting and surface response. If you are handing a generated mesh to an engine, GLB with PBR is a standard hand-off, not a product-specific slogan.

The first stage — especially the prompt stage — also has a research backdrop that is easy to over-read. In DreamFusion: Text-to-3D using 2D Diffusion, Ben Poole, Ajay Jain, Jonathan T. Barron, and Ben Mildenhall used a pretrained 2D text-to-image diffusion model as a prior for text-to-3D synthesis, without 3D training data and without changing the image model. They optimize a 3D representation so that 2D renderings from random angles score well under that prior; the result can be viewed from any angle, relit, or composited into a 3D environment, and the project page notes that the representation can be exported to a mesh for ordinary 3D software. I am citing that as the research problem behind “text in, a viewable 3D asset out.” I am not saying this product implements DreamFusion, a NeRF, or their distillation loss.

A compact pipeline checklist

• Briefing, or a reminder that both inputs exist → the marketing homepage, including the drag-to-rotate gallery.
• Input is a description → the text-to-3D page, up to 1,024 characters, optional Surprise me.
• Input is a photo → the image-to-3D page, PNG/JPG/WEBP, prefer a plain background.
• Detail level → Fast, Standard, Pro, or Ultra, switched independently of the input.
• Hand-off → GLB plus a preview image, PBR included, into the engine or DCC tool you already use.

Keep the brochure at the root. Run the session as input, then density, then a GLB. The mesh is a starting point for a modeler’s judgment, not a replacement for it.