AI Identity Transformation Studio — Parallel GPU Microservice

PIPELINE // INSWAPPER-128 + MOUTH OCCLUSION + CODEFORMER

Studio Portrait & Frame Identity Swap

Upload Target Photo

BUILT-IN TEST PORTRAITS:

SOURCE TARGET FRAME Select or upload a photo

SWAPPED OUTPUT (GPU TENSOR CORES)

-- ms

OPTIMIZATION 1 & 3 // 192×192 EDGE-CLOUD SPLIT + BINARY WEBSOCKET

WS STATE: IDLE

LOCAL RENDER: 0 FPS

GPU TURNAROUND: -- ms

CROP PAYLOAD: ~0.0 KB (192×192)

SWAPPED FRAMES: 0

CLIENT EDGE COMPOSITE (60 FPS + 192×192 CROP BOX) Drag box on canvas to reposition crop ROI

OUTBOUND 192×192 CROP (~8 KB)

INBOUND SWAPPED 192×192 CROP

PHASE 5 // DEEPLABV3 + OPENPOSE + CONTROLNET SD1.5 (DPMSOLVER++ 14-STEP)

Upload Person Photo

1. Select Style Preset (`BODY_STYLE_PRESETS`)

2. Segmentation & Face Preservation Mode

Strength: 0.85

3. Conditioning Prompt (Auto-filled from Preset or Customize)

TEST SUBJECTS:

01 // INPUT SUBJECT

02 // TRANSFORMED OUTPUT

-- s

03 // OPENPOSE SKELETON

04 // DEEPLABV3 WARDROBE MASK

ARCHITECTURE // EDGE-CLOUD GPU MICROSERVICE

WSS DUPLEX STREAM

Continuous Full-Duplex Binary WebSocket. Streams 192×192 JPEG crops (~8 KB) with zero HTTP handshake overhead.

EDGE-CLOUD CROP

Multipart 192×192 face crop swap with SCRFD 320×320 stream detector & precision mouth occlusion mask.

DECOUPLED FULL FRAME

Full-frame identity swap with Nth-frame detection decoupling, mouth preserver, and CodeFormer skin pore restoration.

PHASE 5 INPAINTING

DeepLabV3 GPU segmentation + ControlNet OpenPose + Stable Diffusion 1.5 Inpainting with DPMSolver++ 14-step scheduler.

LIVE ENGINE STATUS JSON (`GET /api/studio/status`)

Loading live telemetry...