ComfyUI FaceSwap SDXL — RunPod Serverless

Character generation with optional face swap, deployed as a RunPod serverless endpoint.

Docker Image

runpod/comfyui-faceswap-sdxl:latest

All models are baked into the image for fast cold starts — no network volume required for inference.

Deploy on RunPod

1. Create a new Serverless Endpoint on runpod.io
2. Set the Docker image to runpod/comfyui-faceswap-sdxl:latest
3. Select a GPU with at least 32 GB VRAM (e.g. A100 40GB, A6000)
4. Optionally attach a network volume mounted at /runpod-volume if you want to persist generated images

API Usage

Request

POST https://api.runpod.ai/v2/{YOUR_ENDPOINT_ID}/runsync
Authorization: Bearer {YOUR_RUNPOD_API_KEY}
Content-Type: application/json

Use /runsync for synchronous requests (waits for the result). For async, use /run and poll /status/{JOB_ID}.

{
  "input": {
    "prompt": "a confident woman in a business suit, professional lighting",
    "negative_prompt": "bad quality, blurry, deformed",
    "width": 832,
    "height": 1216,
    "steps": 35,
    "cfg": 2.0,
    "seed": 42,
    "image_url": "https://example.com/face.png",
    "face_description": "the woman on the left",
    "output": {
      "include_base64": true,
      "save_to_volume": false,
      "volume_path": "outputs"
    }
  }
}

Parameters

Field	Required	Default	Description
prompt	yes	—	Character description
negative_prompt	no	""	What to avoid
width	no	832	Image width
height	no	1216	Image height
steps	no	35	Sampling steps
cfg	no	2.0	CFG scale
seed	no	random	Reproducible seed
image_url	no	—	Face reference URL (enables face swap)
face_description	no	""	Which face to pick from reference
output.include_base64	no	true	Return base64 PNG
output.save_to_volume	no	false	Save to network volume
output.volume_path	no	"outputs"	Folder on /runpod-volume/

Modes

• Text only: Omit image_url — generates character from prompt using CyberRealistic XL
• Face swap: Provide image_url — generates character, then applies face from reference using IPAdapter + InstantID
• Face picking: Provide face_description alongside image_url — uses a vision model to pick the matching face from multi-face scenes instead of defaulting to the largest face

Response

{
  "delayTime": 102,
  "executionTime": 4019,
  "id": "sync-12e21de8-...",
  "output": {
    "duration_seconds": 3.42,
    "image_base64": "<base64-encoded PNG>",
    "image_path": "/runpod-volume/outputs/42_1773510473.png",
    "seed": 42,
    "status": "success"
  },
  "status": "COMPLETED"
}

Resolution Presets

Name	Dimensions
portrait	832 x 1216
landscape	1216 x 832
square	1024 x 1024
small-portrait	640 x 960
small-landscape	960 x 640

Models

All models are baked into the Docker image at build time for fast cold starts.

Model	Purpose	Size
CyberRealistic XL v7	Primary SDXL generation	~6.5 GB
Juggernaut XI	InstantID face refinement	~6.5 GB
CLIP-ViT-H-14	Vision encoding for IPAdapter	~3.9 GB
IPAdapter FaceID Plus v2 SDXL	Face identity transfer	~0.8 GB
IPAdapter Plus Face SDXL	Face style transfer	~0.8 GB
InstantID ControlNet + Adapter	Face structure preservation	~1.7 GB
SAM ViT-B	Face segmentation masks	~0.4 GB
InsightFace (antelopev2 + buffalo_l)	Face detection/recognition	~0.3 GB
YOLOv8m Face	Face bounding box detection	~0.05 GB
Qwen2.5-VL-3B Q4_K_M (GGUF)	Vision-based face picking	~2.3 GB VRAM

GPU Requirements

• Startup: ~21.7 GB VRAM
• Peak (face swap): ~29.5 GB VRAM
• Minimum GPU VRAM: 32 GB

Benchmark

A benchmark script is included to test endpoint performance.

pip install requests
export RUNPOD_API_KEY=your_key

The script uses the endpoint ID defined in benchmark.py — update the ENDPOINT_ID variable to match your endpoint before running.

# Text + face swap, multiple resolutions
python benchmark.py \
  --count 20 \
  --modes text,face \
  --resolutions portrait square \
  --face-url https://example.com/face.png \
  --seed 42

# Text-only
python benchmark.py \
  --count 10 \
  --modes text \
  --resolutions portrait

# Face swap only
python benchmark.py \
  --count 10 \
  --modes face \
  --resolutions portrait \
  --face-url https://example.com/face.png

Images are saved to benchmark_results/<timestamp>/ alongside a report.json with full timing data. The first request runs sequentially to capture cold start time; remaining requests run concurrently.

Flag	Default	Description
--count	10	Total requests
--modes	text	Comma-separated: text, face
--resolutions	portrait	Space-separated: portrait, landscape, square, small-portrait, small-landscape
--face-url	—	Face image URL (required for face mode)
--face-description	—	Which face to pick from reference
--steps	35	Sampling steps
--cfg	2.0	CFG scale
--seed	random	Base seed (incremented per request)
--concurrency	3	Max parallel requests
--output-dir	benchmark_results	Output directory
--timeout	600	Per-job timeout (seconds)