SMD-Qwen-Image-Edit-2511

4-step SMD for instruction-guided image editing. The student adapters, standalone inference code and reproducible examples are included. Upstream base components are downloaded separately.

Training code · SMD collection

Example

Input and four-step edited output, displayed at equal height

Original input · Original edited image

Prompt: Keep her gold-framed sunglasses low on her nose, hairstyle, outfit and background unchanged. Have her hold a single sheet of white paper at chest level with both hands, with large black uppercase "SMD" on it. Match the original soft anime coloring.

Seed: 42. Size: 768 × 1344.

Case settings · Generation receipt

Models

Variant Weights Steps Size
default LoRA 4 1.93 GB

Run

Base: Qwen/Qwen-Image-Edit-2511.

Use Linux, Python 3.12 and an NVIDIA GPU with CUDA support. The included inference code does not import the SMD training repository.

python3.12 -m venv .venv
source .venv/bin/activate
python -m pip install huggingface-hub
# Sign in to Hugging Face.
hf auth login
hf download zjwfufu/SMD-Qwen-Image-Edit-2511 --local-dir SMD-Qwen-Image-Edit-2511 --exclude 'lora/lora.safetensors'
cd SMD-Qwen-Image-Edit-2511
python -m pip install -r requirements.txt
python infer.py --output outputs

The example was run on an A100 80 GB. Model CPU offload reduces GPU memory usage; enough host RAM is still needed to load the upstream components. Use --prompt, --seed and --output to generate another case. --base accepts a local upstream snapshot.

Pass --image path/to/input.png with an editing instruction in --prompt to edit another image.

The first run downloads the upstream base components to the Hugging Face cache. Plan disk space for those components as well as the released weights; subsequent runs reuse the cache.

Weight layout

Student weights are also published at the repository root (smd_lora.safetensors). The legacy weight paths remain available for existing integrations. The download command above excludes the duplicate legacy weights; the supplied loader prefers root weights and falls back to legacy paths. Adapter configurations, merge precision and sampling schedules are preserved.

Sampling

The released sampler uses raw final-segment scale 0.5. The raw grid is built before the model-specific time shift. The included runner follows sampling.json; changing only an upstream pipeline's step count does not reproduce the student.

External classifier-free guidance is disabled for the students. Any embedded guidance remains specified in sampling.json.

The supplied loader validates and applies all LoRA matrices and saved replacement tensors, with the recorded merge precision. Omitting replacement tensors or using a different base can change the result.

License and attribution

Base components retain the terms described in the Qwen/Qwen-Image-Edit-2511 model card. Original code notices and bundled licenses are retained. Adapter use remains subject to the applicable base-model terms.

Downloads last month
80
Inference Providers NEW

Model tree for zjwfufu/SMD-Qwen-Image-Edit-2511

Adapter
(215)
this model

Collection including zjwfufu/SMD-Qwen-Image-Edit-2511