Downloads · 30 days
57
32% of all-time downloads
Merserk/Qwen-Image-VAE-Sharp
Qwen-Image-VAE-Sharp is a text-to-image model from Merserk. Use it when you need an image from a text prompt. The card lists the license as apache-2.0.
<table align="center" <tr <td align="center" <img src="https://cdn-uploads.huggingface.co/production/uploads/63f9610670d45efb75a30a19/DE7j2WGFHjRKSizUjM5rQ.png" alt="Qwen Image VAE Sharp" width="400" / </td <td align=…
Downloads · 30 days
57
32% of all-time downloads
All-time downloads
179
Public
Repo size
2.5 GB
Likes
9
Public
Click a slice to open those files.
.safetensors2.3 GB · 100%
From the Hugging Face model README
Sharper details. Stronger micro-contrast. Cleaner final output.
Qwen Image VAE Sharp, Sharp Plus, and Sharp Ultra are enhanced variants of the standard Qwen Image VAE, tuned for stronger fine detail, cleaner edges, and improved micro-contrast.
All three variants are available in BF16 and FP32.
A color-shift-free VAE that preserves the original image quality while slightly improving sharpness and micro-contrast detail.
Choose Sharp when accurate color reproduction and a conservative enhancement are the priority.
A stable, quality-focused VAE that delivers an optimal balance of sharpness, fine detail, edge clarity, and natural image quality.
Choose Sharp Plus for a balanced everyday option with stronger detail and clean edges without an overly sharpened appearance.
An aggressive sharpening VAE that significantly improves:
Choose Sharp Ultra when maximum visible sharpness and edge definition are preferred.
The Sharp variants deliver crisper, more detailed reconstructions while maintaining the original colors, composition, and overall look.
They are especially well suited for:
sharp_vae_bf16.safetensorssharp_vae_fp32.safetensorssharp_vae_plus_bf16.safetensorssharp_vae_plus_fp32.safetensorssharp_vae_ultra_bf16.safetensorssharp_vae_ultra_fp32.safetensorsFor FP32 VAE decoding, launch ComfyUI with the --fp32-vae flag:
python main.py --fp32-vae
Early access and project support: