Overview

BennyDaBall/Qwen3-4b-Z-Image-Turbo-AbliteratedV1 is an abliterated version of the Tongyi-MAI/Z-Image-Turbo text encoder. Developed by BennyDaBall, this model has been specifically engineered to drastically reduce content refusal rates, particularly for image generation prompts and general content restrictions.

Key Capabilities

• Significantly Reduced Refusals: Achieves a refusal rate of only 4/100 in torture tests, targeting both image generation and general content refusals.
• High Fidelity: Maintains a very low KL Divergence of 0.0004 from the base model, indicating minimal impact on the original model's quality and capabilities despite the ablation.
• Flexible Quantization: Provided in multiple GGUF formats, including F16 (8.05 GB), Q8_0 (4.28 GB), Q6_K (3.31 GB), Q5_K_M (2.89 GB), Q4_K_M (2.50 GB), Q4_K_S (2.38 GB), Q3_K_M (2.08 GB), and Q2_K (1.67 GB), allowing users to choose based on their hardware and performance needs.

Good For

• Users seeking a text encoder with minimal content moderation and refusal behaviors.
• Applications requiring unrestricted image generation prompts.
• Developers who need a model that will generate desired content without frequent "I can't generate that" responses.