Qwen-Image-Edit-2511 - SGLang Documentation

1. Model Introduction

Qwen-Image-Edit-2511 is the 20B editing counterpart to Qwen-Image. It is strongest at changing text, materials, lighting, viewpoint, or composition while reducing drift in regions that were not requested to change. Choose it for identity-sensitive portrait edits, multi-person composition, typography replacement, and geometry-aware design work. It is substantially heavier than small specialist editors, and consistency is improved rather than guaranteed; evaluate untouched-region drift on the actual editing workload.

2. SGLang-diffusion Installation

SGLang-diffusion offers multiple installation methods. You can choose the most suitable installation method based on your hardware platform and requirements. Please refer to the official SGLang-diffusion installation guide for installation instructions.

3. Model Deployment

This section provides deployment configurations optimized for different hardware platforms and use cases.

3.1 Basic Configuration

Qwen-Image-Edit-2511 is a 20B parameter model optimized for image editing tasks. The recommended launch configurations vary by hardware. Interactive Command Generator: Use the configuration selector below to automatically generate the appropriate deployment command for your hardware platform.

3.2 Configuration Tips

See Performance Optimization for acceleration features and their runtime requirements.

--vae-path: Path to a custom VAE model or HuggingFace model ID (e.g., fal/FLUX.2-Tiny-AutoEncoder). If not specified, the VAE will be loaded from the main model path.
--num-gpus: Number of GPUs to use
--tp-size: Tensor parallelism size (only for the encoder; should not be larger than 1 if text encoder offload is enabled, as layer-wise offload plus prefetch is faster)
--sp-degree: Sequence parallelism size (typically should match the number of GPUs)
--ulysses-degree: The degree of DeepSpeed-Ulysses-style SP in USP
--ring-degree: The degree of ring attention-style SP in USP

3.3 Decompose an image into layers on H200

Qwen/Qwen-Image-Layered returns separate RGBA images. For this model, --num-frames 4 requests four output layers. The CLI saves all four PNGs, and DiffGenerator.generate() returns one result per layer. On Linux with NVIDIA CUDA and two H200 GPUs, you can run the conditional and unconditional branches on separate GPUs:

Command

CUDA_VISIBLE_DEVICES=0,1 sglang generate \
  --model-path Qwen/Qwen-Image-Layered \
  --num-gpus 2 \
  --cfg-parallel-size 2 \
  --tp-size 1 \
  --ulysses-degree 1 \
  --quality lossless \
  --enable-torch-compile false \
  --warmup-mode request \
  --width 640 --height 640 --num-frames 4 \
  --num-inference-steps 50 --guidance-scale 4.0 --seed 42 \
  --image-path https://raw.githubusercontent.com/QwenLM/Qwen-Image-Layered/main/assets/test_images/4.png \
  --prompt "a high quality, cute halloween themed illustration, consistent style and lighting" \
  --output-path outputs/qwen-layered \
  --save-output

For a single H200, set CUDA_VISIBLE_DEVICES=0, --num-gpus 1, and --cfg-parallel-size 1. Both configurations use eager execution. Layered does not currently support breakable CUDA graph; enabling BCG falls back to eager execution. The Layered CFG policy gathers the branch predictions before applying the single-GPU arithmetic order, preserving BF16 rounding and alpha values in the validated fixed-seed example. Each GPU still holds a full DiT replica, so CFG parallelism reduces request latency without reducing the model memory needed on each GPU.

4. API Usage

For complete API documentation, please refer to the official API usage guide.

4.1 Edit an Image

Example

import base64
from openai import OpenAI

client = OpenAI(api_key="EMPTY", base_url="http://localhost:3000/v1")

response = client.images.edit(
    model="Qwen/Qwen-Image-Edit-2511",
    image=open("input.png", "rb"),
    prompt="Change the color of the taxi to black.",
    n=1,
    response_format="b64_json",
)

# Save the edited image
image_bytes = base64.b64decode(response.data[0].b64_json)
with open("output.png", "wb") as f:
    f.write(image_bytes)

4.2 Advanced Usage

4.2.1 Cache-DiT Acceleration

SGLang integrates Cache-DiT, a caching acceleration engine for Diffusion Transformers (DiT), to achieve up to 7.4x inference speedup with minimal quality loss. You can set SGLANG_CACHE_DIT_ENABLED=True to enable it. For more details, please refer to the SGLang Cache-DiT documentation. Basic Usage

Command

SGLANG_CACHE_DIT_ENABLED=true sglang serve --model-path Qwen/Qwen-Image-Edit-2511

Advanced Usage

DBCache Parameters: DBCache controls block-level caching behavior:

Parameter	Env Variable	Default	Description
Fn	`SGLANG_CACHE_DIT_FN`	1	Number of first blocks to always compute
Bn	`SGLANG_CACHE_DIT_BN`	0	Number of last blocks to always compute
W	`SGLANG_CACHE_DIT_WARMUP`	4	Warmup steps before caching starts
R	`SGLANG_CACHE_DIT_RDT`	0.24	Residual difference threshold
MC	`SGLANG_CACHE_DIT_MC`	3	Maximum continuous cached steps

TaylorSeer Configuration: TaylorSeer improves caching accuracy using Taylor expansion:

Parameter	Env Variable	Default	Description
Enable	`SGLANG_CACHE_DIT_TAYLORSEER`	false	Enable TaylorSeer calibrator
Order	`SGLANG_CACHE_DIT_TS_ORDER`	1	Taylor expansion order (1 or 2)

Combined Configuration Example:

Command

SGLANG_CACHE_DIT_ENABLED=true \
SGLANG_CACHE_DIT_FN=2 \
SGLANG_CACHE_DIT_BN=1 \
SGLANG_CACHE_DIT_WARMUP=4 \
SGLANG_CACHE_DIT_RDT=0.4 \
SGLANG_CACHE_DIT_MC=4 \
SGLANG_CACHE_DIT_TAYLORSEER=true \
SGLANG_CACHE_DIT_TS_ORDER=2 \
sglang serve --model-path Qwen/Qwen-Image-Edit-2511

4.2.2 CPU Offload

--dit-cpu-offload: Use CPU offload for DiT inference. Enable if run out of memory.
--text-encoder-cpu-offload: Use CPU offload for text encoder inference.
--image-encoder-cpu-offload: Use CPU offload for image encoder inference.
--vae-cpu-offload: Use CPU offload for VAE.
--pin-cpu-memory: Pin memory for CPU offload. Only added as a temp workaround if it throws “CUDA error: invalid argument”.

4.2.3 Known LoRA examples

Use --lora-path at startup or the LoRA management API to load an adapter. Known Qwen-Image-Edit examples include:

5. Benchmark

Test Environment:

Hardware: NVIDIA B200 GPU (1x)
Model: Qwen/Qwen-Image-Edit-2511
sglang diffusion version: 0.5.6.post2

5.1 Speedup Benchmark

5.1.1 Edit a image

Server Command:

Command

sglang serve --model-path Qwen/Qwen-Image-Edit-2511 --port 30000

Benchmark Command:

Command

python3 -m sglang.multimodal_gen.benchmarks.bench_serving \
    --dataset vbench --task image-to-image --num-prompts 1 --max-concurrency 1

Result:

Output

================= Serving Benchmark Result =================
Model:                                   Qwen/Qwen-Image-Edit-2511
Dataset:                                 vbench
Task:                                    image-to-image
--------------------------------------------------
Benchmark duration (s):                  35.31
Request rate:                            inf
Max request concurrency:                 1
Successful requests:                     1/1
--------------------------------------------------
Request throughput (req/s):              0.03
Latency Mean (s):                        35.3053
Latency Median (s):                      35.3053
Latency P99 (s):                         35.3053
--------------------------------------------------
Peak Memory Max (MB):                    47959.35
Peak Memory Mean (MB):                   47959.35
Peak Memory Median (MB):                 47959.35
============================================================

5.1.2 Edit a image with high concurrency

Benchmark Command:

Command

python3 -m sglang.multimodal_gen.benchmarks.bench_serving \
    --dataset vbench --task image-to-image --num-prompts 20 --max-concurrency 20

Result:

Output

================= Serving Benchmark Result =================
Model:                                   Qwen/Qwen-Image-Edit-2511
Dataset:                                 vbench
Task:                                    image-to-image
--------------------------------------------------
Benchmark duration (s):                  286.11
Request rate:                            inf
Max request concurrency:                 20
Successful requests:                     20/20
--------------------------------------------------
Request throughput (req/s):              0.07
Latency Mean (s):                        150.0428
Latency Median (s):                      150.0600
Latency P99 (s):                         283.3843
--------------------------------------------------
Peak Memory Max (MB):                    47971.82
Peak Memory Mean (MB):                   47971.49
Peak Memory Median (MB):                 47971.29
============================================================

​1. Model Introduction

​2. SGLang-diffusion Installation

​3. Model Deployment

​3.1 Basic Configuration

​3.2 Configuration Tips

​3.3 Decompose an image into layers on H200

​4. API Usage

​4.1 Edit an Image

​4.2 Advanced Usage

​4.2.1 Cache-DiT Acceleration

​4.2.2 CPU Offload

​4.2.3 Known LoRA examples

​5. Benchmark

​5.1 Speedup Benchmark

​5.1.1 Edit a image

​5.1.2 Edit a image with high concurrency

1. Model Introduction

2. SGLang-diffusion Installation

3. Model Deployment

3.1 Basic Configuration

3.2 Configuration Tips

3.3 Decompose an image into layers on H200

4. API Usage

4.1 Edit an Image

4.2 Advanced Usage

4.2.1 Cache-DiT Acceleration

4.2.2 CPU Offload

4.2.3 Known LoRA examples

5. Benchmark

5.1 Speedup Benchmark

5.1.1 Edit a image

5.1.2 Edit a image with high concurrency