AI 3D generation is most useful when it removes a known production bottleneck. It is least useful when a team generates hundreds of objects before deciding their scale, style, performance budget, or gameplay function.
This guide follows one asset from brief to Meshy 6, through Blender when necessary, and into Unity with a measurable acceptance test. It extends the broader tool-selection guidance in our overview of the best AI for game models.
Quick read
Key takeaways
- Write an asset contract—scale, camera distance, polygon budget, materials, animation, and export format—before generating.
- Use Image to 3D when visual identity matters and Text to 3D when exploring forms without a settled reference.
- Inspect geometry before spending time on PBR textures, rigging, or animation.
- A successful Unity import proves file compatibility, not production readiness.
- Use Blender and automated validators for topology, transforms, pivots, UVs, LODs, collision, and repeatable export.
What Meshy 6 changes for game assets
Meshy’s current documentation positions Meshy 6 for final-asset quality, while Smart Topology is optimized for game development and real-time rendering. The platform supports text, images, and conversational prompts, followed by texturing, remeshing, rigging, animation, and export.
That is a more complete pipeline than a raw text-to-mesh demo, but every output is still conditional on the input, subject, camera, target device, and quality bar. A background prop and a close-up deforming character should not pass the same review.
| Asset level | Purpose | Minimum proof |
|---|---|---|
| Concept | Choose form and direction | Readable silhouette from key angle |
| Prototype | Test scale and gameplay | Imports, collides, and meets rough budget |
| Production prop | Ship in repeated scenes | Materials, LOD, collision, naming, performance |
| Production character | Animate near camera | Topology, rig, weights, clips, deformation tests |
Define the asset contract first
Write what the game needs, not only what the object looks like. A strong brief says: “Stylized forest shrine, 2.2 meters tall, readable from 15 meters, mobile third-person game, under 12,000 triangles at LOD0, one 2K PBR material set, pivot at ground center, simple box collider, no animation, exported as FBX.”
The contract creates a rejection rule. If the silhouette is wrong, regenerate. If the silhouette is right but the pivot or topology is wrong, fix it downstream. Without that distinction, creators waste credits trying to solve DCC problems with generation prompts.
- Gameplay purpose and expected screen size
- Real-world dimensions, origin, and forward axis
- Triangle budget and LOD targets
- Material channels and texture resolution
- Static, rigged, or animated behavior
- Collision and navigation requirements
- Export format and engine render pipeline
Choose Text to 3D or Image to 3D
Meshy’s official quick start recommends Image to 3D when a visual reference exists and Text to 3D when the creator has only an idea. For game production, the choice is about control: a clean reference anchors silhouette and style; text leaves more room for exploration.
Use a clean subject with minimal background and enough visible structure to infer depth. When consistency matters, prepare front, side, and rear references or a coherent multi-view sheet. Avoid dramatic perspective, hidden limbs, motion blur, reflective clutter, and cast shadows that may be interpreted as geometry or texture.
Use a game-asset prompt framework
A useful prompt names the object, silhouette, proportions, construction, material, style, wear, and exclusions. For example: “Compact stylized treasure chest, broad rectangular silhouette, oversized iron straps, chunky hinges, hand-painted wood and dark metal, light edge wear, closed lid, isolated object, no floor, no floating parts, no text.”
Keep abstract lore out of the geometry prompt unless it changes visible construction. “Cursed by a forgotten king” is not actionable; “thin violet light leaking through three carved seams” is. Generate a small set, compare silhouettes at thumbnail size, and select one direction before texturing.
Review geometry before texturing
Rotate the untextured mesh under neutral lighting. Look for floating pieces, fused gaps, paper-thin surfaces, asymmetry, holes, noisy contours, excessive density, and geometry that only works from the input camera. If the main volume is incorrect, regenerate rather than repairing the wrong idea.
For the selected mesh, compare the triangle budget with the asset contract. Test whether Smart Topology or remeshing preserves the silhouette and important hard edges. A lower number is not automatically better if it breaks deformation, creates shading artifacts, or erases the features players use to recognize the object.
Validate PBR textures in the target pipeline
Check Base Color without lighting baked into it, Normal orientation, Metallic separation, Roughness range, Emission intent, UV seams, and texel density. A sharp 8K texture can still be the wrong asset for mobile hardware or a small background prop.
Create an engine test scene with neutral daylight, a darker interior light, and the actual render pipeline. An asset that looks correct in a web viewer may expose flipped channels, overly glossy surfaces, compression artifacts, or inconsistent color once it enters Unity URP or HDRP.
Move the asset into Unity
Choose GLB for a compact self-contained static exchange when the project has a compatible importer. Choose FBX for conventional Unity character, rig, and animation workflows. After import, place the source under a predictable folder, create a prefab, map materials, establish scale, and keep generated output separate from hand-authored overrides.
Test the asset in a representative scene. Confirm forward direction, ground contact, material pipeline, texture compression, collider behavior, lightmap or probe settings, and frame cost when the expected number of copies is visible. “It appears in the Project window” is not the acceptance criterion.
Use Blender where generation stops
Meshy provides a Blender Bridge that can transfer generated assets into a Blender scene. Blender remains the controlled workspace for applying transforms, fixing normals, adjusting pivots, editing UVs, retopologizing, painting weights, creating LODs, authoring collision, and exporting with a team preset.
Static props may need only scale, pivot, material, and collider cleanup. Deforming characters deserve a stricter path: inspect edge flow around joints, verify bone naming, test extreme poses, repair weights, decide root motion, and validate every required animation in Unity.
Automate a repeatable validation report
A Blender Python validator can report object names, dimensions, unapplied transforms, triangle counts, non-manifold geometry, empty UV layers, material slots, missing image files, bone counts, and export readiness. It should report failures before changing the scene and save corrected output to a new file.
Coding agents can write this script, but the asset contract supplies the truth. Run the validator on one known-good asset and several known failures. If it cannot distinguish them, improve the checks before batch-processing a library.
| Validation | Pass condition | Failure response |
|---|---|---|
| Scale | Matches contract in meters | Apply controlled transform |
| Geometry | Within budget; no critical non-manifold areas | Regenerate or repair |
| Materials | Expected slots and maps resolve | Relink and standardize |
| Rig | Required bones and stable deformation | Re-rig or repaint weights |
| Export | Preset produces repeatable import | Block delivery and report |
The game-ready acceptance checklist
A game-ready asset passes inside the actual game, not inside the generator. Review it at the intended camera distance, under final lighting, in motion, beside neighboring assets, with the expected number of copies, on target hardware.
- Visual: silhouette, style, scale, and materials match the art direction.
- Geometry: topology and triangle density support the asset’s use.
- Animation: rig, weights, clips, and root behavior survive extreme tests.
- Collision: physics is simple, stable, and appropriate for play.
- Performance: LOD, textures, materials, and instance count fit the budget.
- Provenance: source references, generation record, edits, and license context are archived.
- Integration: prefab, naming, folders, import preset, and validation are reproducible.
Frequently asked questions
Are Meshy 6 models ready to use directly in Unity?
They can be imported, but production readiness depends on the asset contract. Check scale, orientation, topology, textures, materials, pivots, animation, collision, LODs, performance, and provenance inside the target Unity project.
Should I export Meshy assets as GLB or FBX?
GLB is convenient for self-contained static assets when the project has a suitable importer. FBX remains a common choice for Unity characters, skeletons, and animation clips. Test the exact importer and render pipeline before standardizing.
Does Meshy 6 create low-poly game assets?
Meshy offers workflows aimed at real-time assets, including Smart Topology and remeshing. You still need to verify the triangle distribution and silhouette against the target camera, hardware, animation, and instance count.
Do I still need Blender after Meshy?
Often, yes. Blender provides precise control over transforms, pivots, topology, UVs, weights, LODs, collision, naming, and export presets. Simple background props may require less cleanup than characters or hero assets.
Is Text to 3D or Image to 3D better for games?
Use Text to 3D for form exploration when art direction is still open. Use Image to 3D when a clean concept or approved design should anchor silhouette and style. The quality of the reference strongly influences the result.
Can Meshy rig and animate a character?
Meshy supports rigging and animation workflows for supported character types. Test bone placement, naming, weights, extreme poses, root motion, and every required clip before accepting the character for gameplay.
How large should game-asset textures be?
Choose resolution from screen size, camera distance, platform memory, compression, and the number of visible assets. A background prop rarely benefits from the same texture budget as a close-up character, even if a higher-resolution source is available.
Can a coding agent automate the Meshy-to-Unity pipeline?
It can automate validators, naming, Blender scripts, batch exports, Unity import presets, and reports when the tools are accessible. Keep approval gates around destructive edits and verify results on known-good and known-bad assets before scaling the automation.
Sources and further reading
- Meshy official documentation
Current platform, model positioning, export formats, plugins, and API overview.
- Meshy quick start
Official generation, inspection, post-processing, and export walkthrough.
- Meshy game asset workflow
Official guidance for generation, remeshing, PBR texturing, rigging, and export.
- Meshy for Blender
Official Blender plugin and Bridge documentation.
- Blender manual: retopology
Official Blender guidance on automatic remeshing and manual topology work.
Next step
