3D
TripoSG 圖片轉 3D
將一張相片轉為可下載的 3D 網格,手機上也能做到。
TripoSG by VAST-AI-Research; DINOv2 by Meta AI; U2-Net by Xuebin Qin and contributors; rembg by Daniel Gatis
Real-ESRGAN by Xintao Wang
4x-UltraSharp by Kim2091, CC BY-NC-SA 4.0; converted to ONNX by Mesh, unchanged learned tensors.
模型授權
TripoSG DiT and VAE ONNX derivatives · MIT
Preserve the complete MIT terms, copyright notice and disclaimer, and identify modifications. No upstream endorsement is granted.
DINOv2 Large conditioning image encoder · Apache-2.0
Preserve the complete Apache-2.0 terms, copyright notices and disclaimers, identify modifications, and respect Apache patent/trademark provisions. No upstream endorsement is granted.
U2-Net matte, rembg ONNX release · Apache-2.0 AND MIT
Preserve the complete applicable MIT/Apache-2.0 terms, copyright notices and disclaimers, identify modifications, and respect Apache patent/trademark provisions. No upstream endorsement is granted.
Real-ESRGAN x4plus original release ONNX derivative · BSD-3-Clause
Retain the original copyright notice, all three BSD conditions and warranty disclaimer with distributed copies. No endorsement using author/contributor names without permission.
Original 4x-UltraSharp ONNX conversion by Mesh · CC-BY-NC-SA-4.0
NonCommercial only. Preserve attribution, full license and modification notice. ShareAlike applies to adapted licensed material. Private compute only, unmetered, requester and operator acceptance required; no global-network execution. Acceptance does not grant commercial permission.
API 參考
透過 REST 提交
使用您的個人 API 金鑰(msk_…),從您自己的工具為此工作負載建立工作。以這種方式啟動的工作,會在您於 您的裝置 連接的同一批裝置上執行——至少須有一部在線上。
https://cp.meshlabs.cloud/v1/organizations/jobs使用 X-Organization-Api-Key: msk_your_personal_api_key 進行驗證。此標頭名稱與組織金鑰相同;個人金鑰使用同一個端點。
POST https://cp.meshlabs.cloud/v1/organizations/job-inputs/presign,將您的檔案上傳到傳回的 upload_url,然後在 job_spec 中傳入 download_url(例如作為 input_image_url)。請求內文
{
"workload_type": "triposg-1-5b-i2m",
"job_spec": {
"input_image_url": "https://…/your-input.png",
"remove_background": true,
"steps": 50,
"guidance_scale": 7,
"grid_resolution": 192,
"texture": true,
"texture_upscale": false,
"texture_upscaler": "ultrasharp"
}
}job_spec 欄位
| 欄位 | 必填 | 預設值 | 說明 |
|---|---|---|---|
input_image_url | 是 | — | 單一物體的相片或渲染圖。主體會自動裁切出來,裁成正方形並縮放為 224x224,因此單一清晰的主體能得到最佳幾何;雜亂的場景與多個物體會讓裁切出錯。本身已具透明背景的 PNG 會略過裁切步驟,是最可靠的輸入。 |
remove_background | 否 | true | 當圖片本身沒有透明背景時,執行分割模型。若圖片已是白底的乾淨裁切圖,或主體已填滿畫面,請關閉此項。 |
steps | 否 | 50 | 整流流步數。低於約 16 時形狀仍可辨識,但會失去細緻的起伏;高於預設值則不再有變化。 |
guidance_scale | 否 | 7 | 網格貼合參考圖片的程度。 |
seed | 否 | — | 可重現生成結果的隨機種子。 |
grid_resolution | 否 | 192 | 用於擷取表面之場的解析度。成本隨三次方成長:256 的工作量約為預設值的 2.4 倍,但差異大多無法在平滑處理後保留。 |
texture | 否 | true | 將相片作為基礎色貼圖投影到網格上。相片拍到的那一面會很清晰;背面則是最近輪廓顏色的塗抹,因為單張圖片看不到它。若要純灰色網格請關閉,檔案較小;若您打算自己貼圖,這也是正確的選擇。 |
texture_upscale | 否 | false | 在投影之前,先對主體執行 4x 超解析度處理。僅當主體在畫面中很小時才值得這麼做;貼圖是以相片本身的解析度建立,因此大而清晰的相片所含細節已超過貼圖所需,這樣只是把我們自己縮小的結果再放大。當相片夠大時會自動略過,所以保持開啟只會多一次下載。 |
texture_upscaler | 否 | ultrasharp | UltraSharp: preferred detail, private NonCommercial use with terms acceptance. x4plus: BSD-3-Clause alternative. Used only when sharpening a small photo is enabled. |
cURL
curl -X POST https://cp.meshlabs.cloud/v1/organizations/jobs \
-H "X-Organization-Api-Key: msk_your_personal_api_key" \
-H "Content-Type: application/json" \
-d '{
"workload_type": "triposg-1-5b-i2m",
"job_spec": {
"input_image_url": "https://…/your-input.png",
"remove_background": true,
"steps": 50,
"guidance_scale": 7,
"grid_resolution": 192,
"texture": true,
"texture_upscale": false,
"texture_upscaler": "ultrasharp"
}
}'https://cp.meshlabs.cloud/v1/organizations/jobs/{job_id}使用同一個 API 金鑰輪詢,直到 status 進入終止狀態(COMPLETED、FAILED、CANCELED 或 MODERATION_BLOCKED)。啟用內容審核時,工作可能會先處於 MODERATION_PENDING 狀態,之後才進入運算佇列。已完成的工作會在每個輸出上附帶預先簽署的 download_url 值。
建立回應
{
"id": "3fa85f64-5717-4562-b3fc-2c963f66afa6",
"workload_type": "triposg-1-5b-i2m",
"status": "PENDING",
"job_spec_json": {
"input_image_url": "https://…/your-input.png",
"remove_background": true,
"steps": 50,
"guidance_scale": 7,
"grid_resolution": 192,
"texture": true,
"texture_upscale": false,
"texture_upscaler": "ultrasharp"
},
"work_units": [],
"outputs": [],
"created_at": "2026-06-08T12:00:00Z",
"updated_at": "2026-06-08T12:00:00Z"
}完成回應
{
"id": "3fa85f64-5717-4562-b3fc-2c963f66afa6",
"workload_type": "triposg-1-5b-i2m",
"status": "COMPLETED",
"job_spec_json": {
"input_image_url": "https://…/your-input.png",
"remove_background": true,
"steps": 50,
"guidance_scale": 7,
"grid_resolution": 192,
"texture": true,
"texture_upscale": false,
"texture_upscaler": "ultrasharp"
},
"work_units": [
{
"id": "…",
"index": 0,
"status": "COMPLETED"
}
],
"outputs": [
{
"id": "…",
"output_type": "<output type>",
"filename": "<output filename>",
"size_bytes": 524288,
"download_url": "<presigned output download URL>"
}
],
"created_at": "2026-06-08T12:00:00Z",
"updated_at": "<completion timestamp>"
}上傳一張相片,即可下載 3D 網格(GLB)。主體會從背景中裁切出來。顏色取自相片:您拍攝的那一面很清晰,背面較柔和,且沒有金屬度或粗糙度貼圖。在您連接的裝置上免費將圖片轉為 3D,手機也可以。首次執行會下載模型;之後的相片啟動更快。進階選項可關閉裁切與貼圖,並在主體在畫面中太小時開啟較大的貼圖。