Image-to-Image
Diffusers
Safetensors
QwenImageEditPlusPipeline
qwen-image-edit
lora
multi-angle
camera-angles
quantized
nf4
Instructions to use mash2005/Qwen-Image-Edit-2511-MultiAngles-Q4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use mash2005/Qwen-Image-Edit-2511-MultiAngles-Q4 with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline from diffusers.utils import load_image # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("Qwen/Qwen-Image-Edit-2511", dtype=torch.bfloat16, device_map="cuda") pipe.load_lora_weights("mash2005/Qwen-Image-Edit-2511-MultiAngles-Q4") prompt = "Turn this cat into a dog" input_image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/cat.png") image = pipe(image=input_image, prompt=prompt).images[0] - Inference
- Notebooks
- Google Colab
- Kaggle
- Local Apps
- Draw Things
Qwen-Image-Edit-2511 + Multiple Angles LoRA (NF4 Quantized)
This is a merged and quantized version of Qwen/Qwen-Image-Edit-2511 with the fal/Qwen-Image-Edit-2511-Multiple-Angles-LoRA fused directly into the base weights.
The transformer is quantized to NF4 (4-bit) with double quantization using bitsandbytes — equivalent to Q4_K_M.
What this does
Generate novel camera angles and viewpoints from a single image. The LoRA was trained on 3000+ Gaussian Splatting pairs with 96 precise camera poses.
Why this exists
The original base model + LoRA requires loading separately. This merge lets you run it without PEFT at load time, at a fraction of the VRAM cost.
Usage
import torch
from diffusers import QwenImageEditPlusPipeline, QwenImageTransformer2DModel
from transformers import BitsAndBytesConfig
from PIL import Image
quant_config = BitsAndBytesConfig(
load_in_4bit=True,
bnb_4bit_quant_type="nf4",
bnb_4bit_compute_dtype=torch.bfloat16,
bnb_4bit_use_double_quant=True,
)
transformer = QwenImageTransformer2DModel.from_pretrained(
"mash2005/Qwen-Image-Edit-2511-MultiAngles-Q4",
subfolder="transformer",
quantization_config=quant_config,
torch_dtype=torch.bfloat16,
device_map="cuda"
)
pipeline = QwenImageEditPlusPipeline.from_pretrained(
"mash2005/Qwen-Image-Edit-2511-MultiAngles-Q4",
transformer=transformer,
torch_dtype=torch.bfloat16,
)
pipeline.enable_model_cpu_offload()
image = Image.open("your_image.png").convert("RGB")
result = pipeline(
image=[image],
prompt="Show this object from the back view, azimuth 180 degrees",
negative_prompt=" ",
num_inference_steps=30,
guidance_scale=1.0,
true_cfg_scale=4.0,
).images[0]
result.save("output.png")
Prompt format
Use azimuth/elevation language for best results:
"front view, azimuth 0 degrees""back view, azimuth 180 degrees""side view, azimuth 90 degrees""top-down view, elevation 90 degrees"
Credits
- Base model: Qwen/Qwen-Image-Edit-2511
- LoRA: fal/Qwen-Image-Edit-2511-Multiple-Angles-LoRA
- Downloads last month
- 5
Model tree for mash2005/Qwen-Image-Edit-2511-MultiAngles-Q4
Base model
Qwen/Qwen-Image-Edit-2511