Neural4D's 3D Studio runs every generation job through two sequential stages: base mesh synthesis powered by the Direct3D-S2 engine, followed by a separate PBR texture bake. Each stage can surface its own class of errors. Most failures fall into one of five categories: bad input, queue timeout, texture pipeline failure, export restriction, or browser-side cache corruption.
| Error symptom | Most likely cause | First action |
|---|---|---|
| Generation fails immediately after submit | Input image violates size or ratio rules | Ensure resolution >256x256; crop to subject |
| Mesh appears with holes or missing faces | Low-contrast background or complex transparency | Use plain background; describe opaque material in prompt |
| White mesh with no texture | PBR texture step timed out or was skipped | Run the AI Texture tool manually |
| Progress spinner frozen for 3+ minutes | Complex model processing (normal) | Wait patiently. Do not refresh. 3+ mins is normal. |
| Export button greyed out | Session expired or authentication error | Open incognito window or log out/in |
Immediate failures before the progress bar even starts are caused by the input image failing a server-side validation check. The Image to 3D pipeline enforces the following requirements:
| Requirement | Details & fix |
|---|---|
| Aspect ratio | Ratios more extreme than 4:1 or 1:4 are rejected. Crop to a roughly square or standard landscape frame before uploading. |
| Subject coverage | If the subject occupies less than roughly 20% of the frame area, the Direct3D-S2 encoder cannot lock onto it. Zoom in and crop tightly to the object. |
| Resolution | The image resolution is recommended larger than 256 x 256 px. Submissions below this threshold may produce incomplete geometry or fail the validation check outright. |
For 3D printing workflows using SLA or FDM printers, this step matters most. A tightly cropped, high-contrast photo of a furniture component or mechanical part produces a far cleaner watertight mesh than a wide environmental shot where the subject is small.
Mesh gaps occur when the Direct3D-S2 encoder cannot confidently reconstruct geometry from the input signal. Three conditions reliably produce this outcome:
| Root cause | What to do |
|---|---|
| Low-contrast background | A white subject on a white background makes it impossible to locate object boundaries. Switch to a neutral grey or any solid color that contrasts clearly with the subject. |
| Transparent or reflective surfaces | Glass, chrome, and clear acrylic have almost no recoverable depth signal. Describe the material as opaque in your prompt (e.g. "matte white ceramic" instead of "glass vase"). Re-apply the correct texture afterward with AI Texture. |
| Multi-object scenes | When the image contains multiple disconnected objects, the engine may merge or drop individual pieces. Isolate to one subject per generation job. |
If you receive a blank white mesh, it means the texture feature was not enabled during your initial generation. When starting an Image to 3D or Text to 3D job, you can turn on the texture option to generate PBR textures alongside the mesh. A white mesh is not an error; it simply means the model has not been textured yet.
If you already have a generated white mesh, there are two ways to apply textures retroactively:
| Method | How to do it |
|---|---|
| From the model details page | Open the generated model from your library and click the Texture button on its details page. This sends the model directly to the AI Texture tool for processing. |
| From AI Texture directly | Go to AI Texture in the Studio sidebar, select your white mesh from your model library, upload a reference image showing the material you want, then click Generate. Each texture generation costs 10 Power credits. If you need more credits, see Pro plan options. |
A generation that takes longer than 3 minutes is not necessarily stalled. Complex objects with fine surface detail, or jobs submitted at higher quality settings, can require significantly more processing time than a standard submission. The progress indicator may appear static while the engine is still working internally.
A job has likely failed only if the spinner shows no change for an extended period (typically 10 or more minutes) and you receive no completion notification. In that case, follow this recovery sequence:
All users, including free plan users, can generate textures, preview the textured result, and export in all six formats: .OBJ, .FBX, .GLB, .USDZ, .STL, and .BLEND. Free plan results are not marked as "Trial." If the export button appears greyed out or fails, the cause is typically a browser session issue rather than a plan restriction. For higher Power limits and priority processing, see the Pro plan details.
If export fails, try these steps in order:
| Step | What to do |
|---|---|
| 1. Clear browser cache | Open an incognito or private window and attempt the export from there. A corrupted session cache is the most common cause of false export failures on paid accounts. |
| 2. Re-authenticate | Log out and back in if your session token expired mid-session, then retry the export. |
| 3. Try an alternate format | If .GLB fails but .OBJ works, download as OBJ and convert to your target format using Blender's free export pipeline. |
Most of the errors above are avoidable with a short pre-flight checklist before every submission:
| Check | What to verify |
|---|---|
| Image prep | Crop tightly to the subject, remove busy backgrounds, and use an image with a resolution larger than 256 x 256 px. |
| Prompt clarity | Name one specific object and specify the material as opaque. "A ceramic coffee mug with a smooth white glaze, single object, centered" outperforms "a coffee shop table with cups and plates." |
| Power balance | Confirm you have sufficient Power balance before submitting. A job that starts with insufficient credits fails partway through and cannot be auto-recovered. |
| Browser setup | Use Chrome or Edge with ad-blocker extensions disabled. Some aggressive content blockers intercept the WebSocket connection that drives the generation progress tracker. |
If a generation completes but the shape is wrong (incorrect proportions, a missing component, or distorted topology), you do not need to restart from scratch. Neural4D-2.5 is the conversational AI modeling assistant available exclusively for Text to 3D and Image to 3D outputs.
Open the result and type a plain-language correction. For example: "make the legs 20% thinner and extend them downward" or "remove the extra protrusion on the top surface." Neural4D-2.5 applies targeted edits to the mesh without discarding the full geometry, which is faster than a full regeneration and does not require an additional Power deduction for minor shape corrections.
Note: Neural4D-2.5 applies only to 3D generation outputs. It is not available for Text to Image or Text to Video results; for video outputs, adjust the prompt and regenerate.
Still need help? If you encounter other issues, visit the Neural4D Help Center for more detailed guidance pages.