← All models

Stable Point Aware 3D (SPAR3D)

3D Available Full comparison ↗

stability-ai/spar3d · by Stability AI · large-reconstruction-model

Pricing — 1 offering(s)

Stability AI

About this offering: Stability AI's successor to Stable Fast 3D, announced with NVIDIA at CES 2026, with open weights under the Stability AI Community License. A point-cloud stage completes the unseen back of an object and allows real-time mesh edits. It costs 4 credits ($0.04) per generation on Stability's platform, against 10 credits for Stable Fast 3D (Stability pricing page, checked 2026-10-11).

Image-to-3D (point-aware, single shot)

  • $0.01 / credit Current 2026-09-02 → present
    "Stable Point Aware 3D ... 4" credits on the Developer Platform pricing table; 1 credit = $0.01, so $0.04 per…

    "Stable Point Aware 3D ... 4" credits on the Developer Platform pricing table; 1 credit = $0.01, so $0.04 per generation — the cheapest hosted image-to-3D rate currently tracked. Two-stage (point-cloud sampling then meshing) gives complete back-side structure and supports real-time edits; ~0.7s per object.

Showing the active price and any recorded history. Full pricing history is available via the paid API — see API docs.

See how Stable Point Aware 3D (SPAR3D) fits into a cost-aware routing setup

See how →

Capability profile

How Stable Point Aware 3D (SPAR3D) rates across core capability dimensions, with the task-level evidence behind each rating.

geometry quality Moderate

Better structural completeness than SF3D thanks to the point-cloud stage — reconstructs occluded geometry rather than guessing a flat back. Still a speed-first model, not a fidelity leader.

prompt alignment Moderate

Single-image conditioning; the point stage improves plausibility of unseen regions.

texture quality Moderate

Outputs a UV-unwrapped textured GLB directly, as with SF3D.

topology control Weak

No topology controls; real-time point/mesh editing is the differentiator instead.

rigging and animation Weak

Static mesh output.

input flexibility Weak

Single image only (plus optional edited point cloud).

generation speed Strong

~0.7s per object — near-real-time.

Ratings are estimated — limited independent data is available for this model.

Operator guidance

The default pick among the speed-first, lowest-cost hosted models — prefer it over Stable Fast 3D (cheaper and structurally more complete). Step up to Hunyuan3D, Hi3D, TRELLIS 2, or Rodin when geometric detail matters, or to Tripo / Meshy for text input and a full asset pipeline.

Use cases

Limitations

Citations