Hi, how can we help you today?

Advanced 3D Generation Parameters

Generating high-quality 3D models efficiently depends on understanding and tuning a few core parameters: Steps, Guidance, Face Count, and Input Image Quality. Whether you're prototyping quickly or preparing high-fidelity assets, fine-tuning these controls can dramatically improve your results and workflow speed.


Understanding 3D Generation Parameters

Each generation parameter plays a specific role in shaping your 3D model:

  • Steps enhance the quality of detail and texture.

  • Guidance governs how closely the output matches your input image.

  • Face Count (Max Faces) sets the polygon density of your mesh.

  • Input Image quality directly influences the model’s ability to reconstruct accurate geometry.

Mastering these parameters enables you to match your outputs to different goals: rapid iteration, visual style testing, or production-quality assets.


Input Images

You can upload 1 to 10 input images, depending on the selected 3D generation model. For best results:

  • Use high-resolution images with a transparent or plain background.

  • Prefer images with 3D qualities (e.g. lighting, depth cues, realistic proportions).

  • Avoid flat, illustrative artwork—these tend to underperform due to lack of depth.

The more dimensional the image looks, the better the model can reconstruct accurate geometry and volume.


Step Count

Steps control how many denoising iterations the model performs during generation. More steps = more refinement.

Recommended Ranges:

Step Count

Description

10–20

Rapid prototyping with basic detail

30–50

Balanced quality and speed for general use

50+

Maximum detail; diminishing returns begin

💡 Tip: Start with 30 steps for most models. Only increase beyond 50 when details are essential.

Examples:

  • Prototyping (Hunyuan3D): 30/40 steps is a good default. You can reduce further if speed is critical and quality is secondary.

  • Trellis – Stage 1: Use higher step count in Stage 1 to establish strong geometry early, and fewer steps in Stage 2 to avoid hallucinations.


Face Count (Max Faces)

Max Faces determines how many polygons are used to build the mesh. More faces = more geometric detail—but at a performance and file size cost.

Recommended Ranges:

Face Count

Use Case

1K–10K

Prototypes, mobile apps, VR, background assets

10K–40K

Real-time games, most general use cases

40K+

Hero assets, cinematics, close-up rendering

⚠️ Caution: Setting max_faces too high without enough steps might result in dense but poorly formed meshes.
For example, using 1 step and 100 faces is ineffective—you're refining geometry that was never defined properly.


Guidance: Fidelity vs. Flexibility

Guidance controls how strictly the model follows your input image. It balances faithfulness to source with creative 3D interpretation.

Recommended Ranges:

Range

Description

1.0–3.0

Loose adherence, more creative liberty

3.0–7.5

Balanced fidelity and flexibility (recommended default)

7.5–10.0

Very strict adherence to image, may cause distortions

Examples:

  • With Perspective Views: High guidance may introduce visual artifacts as the model tries to rigidly replicate ambiguous geometry.
    → Use medium guidance (5–7.5) for most perspective or ¾ views.

  • For front-facing, symmetrical images: Higher guidance (~8) can help preserve detailed structures more accurately.


Paint Option

Even during early prototyping, texture output can help evaluate visual style.

  • Steps contribute heavily to texture refinement.

  • SDF Resolution + Max Faces may also impact how well textures map to the model geometry.


Was this helpful?