Independent Technical Study · Desktop Web Runtime Validated
An AI-generated backpack, rebuilt for the web.
A study in structure, materials and a more usable 3D asset.
The starting model already worked in a browser. The harder part was separating the fabric, straps and hardware so they could be worked on properly. This study shows that rebuild, and the lighter web version made from it.
Read the study ↓The starting point
Opening the file was only the starting point.
The original model combined the product into one mesh and one material. I rebuilt its structure so different parts could be worked on separately, then prepared a web version of that fuller working model.
What I worked on
- Checked whether the fused mesh and single baked material could be repaired; selective repair did not preserve enough of the asset, so I rebuilt it.
- Organized the rebuild into 59 named production objects with UV layouts and nine PBR materials.
- Worked separately on the fabric, webbing, padding and plastic hardware.
- Grouped the web version into 14 meshes, keeping a mapping back to all 59 parts, and prepared KTX2 textures.
- Compared the web file with the fuller production model and recorded what the desktop checks did and did not cover.
The finished images
The shape of the rebuild.
Up close
Fabric, straps and the small things.
The close-ups show the material work across the bag, from the woven surfaces to the plastic hardware.
More material details
The web version
83.39%
A lighter file from the rebuilt model.
49,266,288-byte production model → 8,183,388-byte web model. This compares the model files, not the total page download or the original AI file.
The original AI file was smaller still at 1,317,164 bytes, and already opened in a browser. This percentage is not a before-and-after comparison with that original file.
A quick walkthrough
Watch the recorded demo.
Recorded study demonstration · 24 seconds
The rendered images and browser version are separate outputs. Lighting and materials can look different.
From a study to your project
What a project like this can include
- Structured production asset and UV layouts
- Authored PBR materials and product-detail visuals
- Optimized desktop web-runtime asset with component mapping
- 24-second study demonstration
- Validation and handoff notes with documented limits
About this study
An independent reconstruction study, not a client project or a manufacturer’s product specification. The validation record includes 0 errors and 54 approved tool-support warnings; it is not a zero-warning asset.
Technical notes
Earlier desktop testing used an AMD Radeon 780M / Ryzen 9 8945HS, Chrome 154, at 1440 × 1000. Stage05 checked the viewer controls; it did not establish a new performance benchmark. Physical phones were not tested.
- Production structure
- 59 semantic objects
- Runtime grouping
- 14 meshes · 59 mapped components
- Runtime geometry
- 39,610 triangles · 9 PBR materials
- Web-runtime file
- 8,183,388 bytes
- glTF validation
- 0 errors · 54 approved tool-support warnings
- Physical mobile devices
- NOT_TESTED
Independent Technical Study, not a client commission, brand affiliation or physical-product certification.
No mobile-optimized, guaranteed-FPS, Shopify or AR delivery claim is made.
Original AI input, reference images, model weights and internal proof are not offered as downloads.
The recorded study and any interactive demo are separate outputs. A recording alone does not establish that an online viewer is available.
- The original AI asset was already browser-loadable and smaller: 1,317,164 bytes.
- The 83.39% reduction compares the 49,266,288-byte production source with its 8,183,388-byte web-runtime derivative.
- glTF validation recorded 0 errors and 54 approved KTX2 tool-support warnings; this is not zero warnings.
- The reconstruction contains authored interpretations, not recovered manufacturer specifications.
The interactive version
Try it yourself.
Turn the backpack around, look closely at its materials, and compare the two study configurations.
- Rotate and zoom
- Compare the depth configurations
- Reset the view
Desktop study. It has not been tested on a physical phone.
Live lighting and materials differ from the rendered images.
Watch the recorded demo