Professional prompt engineering strategies for Hunyuan 3D v3's three generation modes. Parameter optimization, multi-view techniques, and production workflows for game development, e-commerce, and 3D printing.
Hunyuan 3D v3 delivers production-ready 3D models, but output quality depends entirely on understanding its three generation modes and their parameter trade-offs.
Hunyuan 3D v3 compresses the 3D modeling timeline to minutes through a two-stage architecture: Hunyuan3D-DiT generates geometry via flow-based diffusion, while Hunyuan3D-Paint synthesizes high-resolution textures with physically-based rendering properties1. The separation of shape and texture generation allows each stage to optimize independently, producing polygon-efficient meshes suitable for real-time applications.
Understanding how to communicate with this system determines whether you receive a usable production asset or an unusable artifact. The model exposes three distinct input modalities on fal: text-to-3D, image-to-3D, and sketch-to-3D. Each pathway requires different prompt strategies and parameter configurations to achieve optimal results.
The text-to-3D endpoint generates complete 3D models from natural language descriptions. Effective prompts follow a consistent structure that provides the model with sufficient geometric and material context.
Construct prompts with these elements in sequence:
Consider the difference between "a chair" and "mid-century modern armchair with walnut frame, navy velvet upholstery, tapered legs, and brass feet." The latter provides material context, stylistic reference, and structural specificity that guides generation toward a predictable outcome.
A text-to-3D request with error handling:
import { fal } from "@fal-ai/client";
try {
const result = await fal.subscribe("fal-ai/hunyuan3d-v3/text-to-3d", {
input: {
prompt:
"Mid-century modern armchair with walnut frame and navy velvet upholstery",
generate_type: "Normal",
enable_pbr: true,
},
});
console.log(result.model_glb.url); // GLB download URL
console.log(result.model_urls.obj.url); // OBJ download URL
console.log(result.thumbnail.url); // Preview image
} catch (error) {
if (error.status === 422) {
console.error("Invalid input parameters:", error.body);
} else if (error.status === 401) {
console.error("Authentication failed. Check FAL_KEY.");
} else {
console.error("Generation failed:", error.message);
}
}The API exposes parameters that control output characteristics:
| Parameter | Options | Pricing Impact |
|---|---|---|
| generate_type | Normal, LowPoly, Geometry | Normal: $0.375, LowPoly: $0.45, Geometry: $0.225 |
| enable_pbr | true/false | Adds $0.15 when enabled |
| face_count | Custom polygon count | Adds $0.15 for non-default values |
For real-time applications, configure generate_type: "LowPoly" with reduced face count. For visualization or 3D printing, use generate_type: "Normal" with enable_pbr: true.
Image-to-3D converts photographs into textured meshes. Input image quality directly constrains output fidelity.
Optimal input images share these characteristics:
The model supports optional multi-view input using back, left, and right images alongside the primary front view. This adds $0.15 to the generation cost but reduces reconstruction ambiguity for complex geometry. Single-view input works adequately for geometrically simple objects.
The sketch-to-3D mode combines line drawings with text prompts, creating an intuitive pathway from concept to dimensional asset.
Prepare sketches with these considerations:
The sketch establishes geometric structure. The accompanying text prompt specifies attributes not visible in the drawing: materials, surface finishes, and color. Avoid restating structural information already present in the sketch. Instead, focus the prompt on material qualities like "leather upholstery, chrome metal frame, glossy finish."
The generate_type parameter controls output style and cost:
Normal ($0.375): Fully textured models with color and surface detail. Use for visualization, e-commerce, and assets where appearance matters.
LowPoly ($0.45): Reduced polygon count while preserving textures. Optimized for real-time rendering in game engines. Despite higher cost, the polygon efficiency saves downstream processing.
Geometry ($0.225): Untextured white meshes for downstream processing. Use when applying custom materials in external software or when only shape matters.
Enabling PBR adds metalness, roughness, and normal maps to the output for $0.15 additional cost. These maps are essential for photorealistic rendering in modern pipelines like Unreal Engine and Unity HDRP. Skip PBR when targeting stylized aesthetics or when custom shaders will replace generated materials.
Different applications require distinct optimization strategies:
Game development: Set generate_type: "LowPoly" for real-time rendering compatibility. Disable PBR if implementing custom shader systems. Total cost: $0.45.
E-commerce visualization: Use generate_type: "Normal" with enable_pbr: true. Provide multiple product photographs when using image-to-3D for best reconstruction. Total cost: $0.375 + $0.15 = $0.525.
3D printing: Configure generate_type: "Normal" or "Geometry". PBR is unnecessary since physical materials replace textures. Export OBJ for compatibility with mesh repair utilities. Total cost: $0.225 to $0.375.
Architectural visualization: Leverage sketch-to-3D for rapid concept iteration. Enable PBR for realistic material representation. Total cost: $0.375 + $0.15 = $0.525.
Geometry artifacts: Holes and extrusions typically indicate insufficient polygon budget. Request higher face count (adds $0.15) for complex organic shapes.
Blurry textures: Generic material descriptions produce generic textures. Specify materials precisely: "brushed aluminum" rather than "metal," "rough oak wood" rather than "wood."
Structural misinterpretation: When text prompts fail to communicate complex geometry, switch to sketch-to-3D mode. A simple outline often conveys structure more reliably than elaborate verbal description.
Multiple objects in output: The model performs best on single isolated subjects. Generate complex scenes by creating individual assets and compositing in 3D software.
Inconsistent results: The model incorporates stochastic elements. Generate multiple variations and select the optimal result rather than expecting perfection from any single generation.
The API returns both GLB and OBJ formats. GLB packages geometry, textures, and materials into a single file suitable for web viewers and game engines. OBJ separates geometry from its accompanying MTL material definition, offering better compatibility with traditional 3D editing software. The response also includes a thumbnail preview image.
Both formats import directly into Unity, Unreal Engine, Blender, and Maya. PBR materials integrate with modern rendering pipelines without conversion.
Zhao, Z., et al. "Hunyuan3D 2.0: Scaling Diffusion Models for High Resolution Textured 3D Assets Generation." arXiv:2501.12202, 2025. https://arxiv.org/abs/2501.12202 ↩
A generative media engineer with a focus on growth, Zach has deep expertise in building RAG architecture for complex content systems.
Hundreds of production-ready image, video, and audio models behind one API.
A head-to-head of MiniMax H3 and fal Research's post-trained MiniMax H3 Max on fal, nine identical text, image, and reference-to-video shots, plus specs and pricing.
Complete guide to AI-powered 3D model generation for Blender. Learn how to transform images into production-ready 3D assets with advanced optimization techniques.
Prompt structures for 30-second shots, reference control, native audio, physical continuity, and cleaner motion in Seedance 2.5. All 10 examples include the exact prompt and the generated video.
