kukidevalml/Qwen3-4b-Z-Image-Turbo-AbliteratedV1

TEXT GENERATIONConcurrency Cost:1Model Size:4BQuant:BF16Ctx Length:32kTool Calling:SupportedPublished:Jun 10, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

The kukidevalml/Qwen3-4b-Z-Image-Turbo-AbliteratedV1 is a 4 billion parameter model based on the Qwen3 architecture, specifically an abliterated version of the Z-Image-Turbo text encoder. Developed by kukidevalml, this model is engineered to significantly reduce image generation and general refusal rates, achieving a 0.0004 KL Divergence and only 4/100 refusals in torture tests. It is primarily designed for users seeking a less restrictive model for image generation tasks, ensuring higher compliance with user prompts.

Loading preview...

Qwen3-4b-Z-Image-Turbo-AbliteratedV1 Overview

This model is an "abliterated" version of the Z-Image-Turbo text encoder, built upon the Qwen3-4b architecture. Its core purpose is to drastically minimize content refusal rates, particularly for image generation prompts and general conversational refusals. The developer, kukidevalml, utilized a "p-e-w heretic method" over 1000 trials to achieve this, resulting in a model that aims to generate content as requested by the user.

Key Capabilities

  • Reduced Refusal Rates: Engineered to bypass common refusal mechanisms in both image generation and general text prompts, achieving a reported 4/100 refusal rate in testing.
  • Minimal "Lobotomy" Effect: Maintains a very low KL Divergence of 0.0004, indicating that the ablation process has not significantly altered the model's core knowledge or capabilities beyond reducing refusals.
  • Flexible Content Generation: Designed to be highly compliant with user requests, enabling broader creative freedom in outputs.

Good For

  • Unrestricted Image Prompting: Ideal for users who require a model that will generate images based on diverse and potentially sensitive prompts without frequent refusals.
  • Creative Applications: Suitable for creative writing, roleplay, or any application where prompt adherence and avoidance of content filters are paramount.
  • Exploratory Content Creation: Useful for developers and creators who need a model that prioritates user intent over built-in guardrails, while understanding the associated responsibilities.

Finetunes
1 models