import torch
from diffusers import DiffusionPipeline
# switch to "mps" for apple devices
pipe = DiffusionPipeline.from_pretrained("QuantFunc/Nunchaku-Qwen-Image-2512", dtype=torch.bfloat16, device_map="cuda")
prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
image = pipe(prompt).images[0]License & Attribution
These are quantized derivative weights of Qwen/Qwen-Image-2512 (Qwen-Image-2512).
- Modifications: the original weights were quantized (e.g. W4A4 / FP4 / INT4 / FP8) and repackaged for the QuantFunc inference engine — a "modification" under Apache-2.0 §4(b).
- License: Apache License 2.0 (inherited from the base model), included as
LICENSE. Upstream copyright and attribution notices are retained. - This repository is not affiliated with or endorsed by the upstream model authors.
Disclaimer: "Nunchaku" / "SVDQuant" name the quantization method/format (the open-source SVDQuant work by MIT HAN Lab, Apache-2.0). This repository is an independent re-quantization and is not affiliated with, sponsored by, or endorsed by MIT HAN Lab or the Nunchaku project. Official Nunchaku releases are under the
nunchaku-ai/nunchaku-technamespaces.
QuantFunc
🤗 Hugging Face | 🤖 ModelScope | 💻 GitHub | 💬 WeChat (微信) | 🎮 Discord
⚡ Qwen-Image-2512 — SVDQ (Nunchaku) pre-quantized text-to-image. 2x–11x faster with the QuantFunc plugin; 100% Nunchaku-ComfyUI compatible.
Offline-quantized Qwen-Image-2512 text-to-image checkpoints in best-quality / balanced / ultimate-speed variants (INT4 & FP4), ready to drop straight into ComfyUI.
Powered by the QuantFunc ComfyUI plugin — the fastest diffusion inference engine:
- 🚀 2x–11x speedup over standard BF16/FP16 Python pipelines (pre-exported → even faster loading).
- ⚙️ Native C++/CUDA (
libquantfunc.so/quantfunc.dll) with zero Python model dependencies. - 🧩 Dual engine (SVDQ offline + Lighting runtime 4-bit), zero-cost LoRA stacking, reference-image editing & inpainting.
- 🟢 Full GPU coverage — RTX 20/30/40/50 · A100/H100/H200/B100/B200/GB300 · RTX 6000 Ada / PRO Blackwell (CUDA 12 & 13); native FP4 on Blackwell.
👉 Install the plugin: https://github.com/RealJonathanYip/ComfyUI-QuantFunc
Introduction
We are excited to share our latest model series based on nunchaku + qwen-image-2512 quantization. These models are carefully optimized to maintain high-quality output while significantly improving inference speed and efficiency. All models are 100% compatible with the nunchaku-comfyui && lora plugin and can be used directly in ComfyUI.
Gallery
 |
 |
 |
 |
 |
 |
Model Checkpoints
| Name | low_rank | Notes |
|---|---|---|
| nunchaku_qwen_image_2512_best_quality_fp4 | 256 | Best quality model, suitable for scenarios with extremely high quality requirements |
| nunchaku_qwen_image_2512_best_quality_int4 | 256 | Best quality model, suitable for scenarios with extremely high quality requirements |
| nunchaku_qwen_image_2512_ultimate_speed_int4 | 32 | Ultimate speed model, prioritizing inference speed |
| nunchaku_qwen_image_2512_ultimate_speed_fp4 | 32 | Ultimate speed model, prioritizing inference speed |
| nunchaku_qwen_image_2512_balance_int4 | 128 | Balanced model, achieving the best balance between quality and speed |
| nunchaku_qwen_image_2512_balance_fp4 | 128 | Balanced model, achieving the best balance between quality and speed |
4 steps workflow
Here’s a workflow example of integrating 4-step LoRA in ComfyUI. If you don’t need 4-step LoRA, simply remove the LoRA node.
Coming Soon
If you encounter any issues during use, feel free to join our community for feedback:
- Join our Discord server
- Scan the QR code below to join our WeChat group
We will add support for build in lora and qwen-image-edit-2511 in approximately one month.
- Downloads last month
- 1,373
Model tree for QuantFunc/Nunchaku-Qwen-Image-2512
Base model
Qwen/Qwen-Image-2512