Gemini 3 + riftrunner on LMArena: 3D Fox Modeling vs SOTA

I asked riftrunner (rumored Gemini 3) on LMArena to generate a 3D fox. Then I compared the geometry and textures against other state-of-the-art 3D models.

What I asked for

• Prompt: “high-quality 3D fox, clean topology, readable fur, neutral studio lighting.”


• Output: OBJ/mesh-style preview from riftrunner; side-by-side with two SOTA baselines.


• Environment: LMArena Battle mode, no manual post-editing.

Why this matters

• Shape fidelity is the hard part: ears, snout, and paws often collapse in fast 3D pipelines.


• Mesh cleanliness impacts rigging and animation downstream.


• Lighting neutrality keeps focus on geometry quality rather than stylized shading.

How riftrunner (Gemini 3) performed

• Topology: kept ear thickness and snout volume without fusing edges—a win over Baseline B.


• Proportions: body-to-head ratio felt natural; paws remain distinct (many models blob here).


• Texture read: fur directionality appears, though fine strands are still simplified.


• Artifacts: mild smoothing on tail start; no major mesh tearing observed.

Comparison vs SOTA baselines

1. Baseline A (generalist 3D model): sharper fur but heavier topology noise; paws partially merged.


2. Baseline B (stylized generator): fun colors, weaker anatomy; missing inner-ear depth.


3. riftrunner (Gemini 3): most balanced on anatomy and clean surfaces; texturing is conservative but usable.

Takeaways for 3D teams

• Use riftrunner when you need clean anatomy and riggable meshes with minimal cleanup.


• Layer a texture pass if you need richer fur detail; geometry is already stable.


• For stylized looks, a post-style transfer may beat asking the generator to do both shape and style.

Bottom line

For this fox sample, riftrunner (Gemini 3 on LMArena) delivered the best geometry-to-cleanliness ratio among the tested SOTA models. If more complex poses or quad-based retopo are needed, this is a strong starting point with less manual fixing.