Abiray/Qwen-Image-2.1-Turbo-GGUF

Qwen-Image-2.1-Turbo GGUF

Quantized GGUF weights for Alibaba's official Qwen-Image-2.1-Turbo distilled diffusion model.

These checkpoints enable lightning-fast text-to-image and image-editing generation in ComfyUI at a fraction of the native VRAM requirement, taking full advantage of 4–8 step Turbo sampling.

---

📦 Quantization Breakdown

All models were converted using ComfyUI-GGUF tools and quantized using a patched multi-threaded llama-quantize build natively supporting the 297-tensor qwen_image architecture.

Filename	File Size	Recommended GPU / VRAM	Quality Tier & Notes
qwen_image_2.1_turbo_Q8_0.gguf	7.59 GB	12 GB – 16 GB+	Reference grade; indistinguishable from native BF16.
qwen_image_2.1_turbo_Q6_K.gguf	5.88 GB	10 GB – 12 GB	High-fidelity; preserves complex typography and prompt nuance.
qwen_image_2.1_turbo_Q5_K_M.gguf	5.01 GB	8 GB – 12 GB	Excellent texture detail and micro-contrast.
qwen_image_2.1_turbo_Q5_K_S.gguf	4.94 GB	8 GB – 10 GB	Balanced 5-bit quant; slight memory savings over Q5_K_M.
qwen_image_2.1_turbo_Q5_0.gguf	4.94 GB	8 GB – 10 GB	Standard uniform 5-bit quantization; consistent quality across layers.
qwen_image_2.1_turbo_Q4_K_M.gguf	4.19 GB	6 GB – 8 GB	⭐ Recommended Sweet Spot: Best quality-to-speed ratio.
qwen_image_2.1_turbo_Q4_K_S.gguf	4.06 GB	6 GB – 8 GB	Compact 4-bit k-quant; ideal balance for strict 6 GB/8 GB envelopes.
qwen_image_2.1_turbo_Q4_0.gguf	4.05 GB	6 GB – 8 GB	Standard legacy 4-bit format; clean and fast dequantization.
qwen_image_2.1_turbo_Q3_K_M.gguf	3.19 GB	6 GB VRAM / Laptops	Ultra-compact footprint with strong core composition retention.
qwen_image_2.1_turbo_Q3_K_S.gguf	3.11 GB	4 GB – 6 GB	Lightweight 3-bit quant tailored for tight memory limits.
qwen_image_2.1_turbo_Q2_K.gguf	2.45 GB	4 GB VRAM / Extreme Low-Spec	Minimal footprint; noticeable texture loss, but fits on almost any system.

---

🖼️ Sample Generations (Q4_K_M)

Prompt

Showcase Image 1

Prompt

Showcase Image 2

Prompt

Showcase Image 3

Prompt

Showcase Image 4

Prompt

Showcase Image 5

Prompt

Showcase Image 6

Prompt

Showcase Image 7

Prompt

Showcase Image 8

Prompt

Showcase Image 9

---

⚡ Recommended Generation Parameters

Because this is the distilled Turbo variant, your sampling settings must match the accelerated schedule:

• Sampler: euler
• Scheduler: simple or normal
• Steps: 4 to 8 steps (6 steps is ideal for general fidelity)
• CFG Scale: 1.0

  ⚠️ Important: Do not increase CFG above 1.0 on this distilled checkpoint. Distillation collapses guidance requirements; CFG values > 1.0 will cause harsh contrast clipping and visual artifacts.

• Negative Prompt: Leave empty (ignored when CFG = 1.0).

---

🛠️ ComfyUI Setup & Required Components

1. Install Node

Make sure you have the latest version of ComfyUI-GGUF installed in your ComfyUI/custom_nodes/ directory.

2. File Placement

• Diffusion Model (GGUF): Download your chosen .gguf file from this repo and place it in:
  ComfyUI/models/diffusion_models/ (or ComfyUI/models/unet/)
• Text Encoder: Download qwen3vl_8b_int8_convrot.safetensors or qwen3vl_8b_w4a8.safetensors from Comfy-Org/Qwen-Image-2.1 and place it in:
  ComfyUI/models/text_encoders/
• VAE: Download qwen_image_2.1_vae_bf16.safetensors from Comfy-Org/Qwen-Image-2.1 and place it in:
  ComfyUI/models/vae/

---

🎨 Ready-to-Use Workflows

You can drag and drop either of these pre-built workflow JSONs straight into ComfyUI to start generating:

1. Text-to-Image Workflow:
   Download Text2Image Workflow JSON
2. Image-to-Image / Editing Workflow:
   Download Image2Image Edit Workflow JSON

(Just switch the model loader node to point to your new qwen_image_2.1_turbo_Q*.gguf file and adjust steps to 4–8 with CFG set to 1.0!)

---

📑 Acknowledgements & Credits

• Base model developed and released by the Qwen Team / Alibaba.
• GGUF diffusion support enabled via city96/ComfyUI-GGUF and llama.cpp.