fix(rsplat): use concrete object classes for SAM 3D grounding

SAM 3D uses Grounding DINO which needs real noun classes, not abstract
terms. Use broad multi-class prompt with very low threshold (0.05) to
detect objects in any image type.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
Jeff Emmett 2026-03-16 14:10:32 -07:00
parent 31025c24d7
commit 96fd5ef756
1 changed files with 1 additions and 1 deletions

View File

@ -1035,7 +1035,7 @@ app.post("/api/3d-gen", async (c) => {
Authorization: `Key ${FAL_KEY}`,
"Content-Type": "application/json",
},
body: JSON.stringify({ image_url, prompt: "everything", detection_threshold: 0.2, export_textured_glb: true }),
body: JSON.stringify({ image_url, prompt: "person . animal . object . furniture . vehicle . building", detection_threshold: 0.05, export_textured_glb: true }),
signal: controller.signal,
});
clearTimeout(timeout);