BennyDaBall/Z-Image-Engineer-V6

Hugging Face
TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jun 6, 2026License:apache-2.0Architecture:Transformer0.1K Open Weights Featherless Exclusive Warm

Z-Image-Engineer V6 by BennyDaBall is a 4 billion parameter Qwen-based text encoder fine-tuned for Z-Image workflows, featuring a 32768 token context length. It is optimized for dual-role performance: enhancing minimal seed prompts into rich, structured visual narratives and serving as a direct Z-Image text encoder. This model utilizes a SMART DoRA training system to improve scene composition, lighting, texture, and depth separation in generated image prompts, while also reducing repetitive output.

Loading preview...

Z-Image-Engineer V6 (4B)

Z-Image-Engineer V6, developed by BennyDaBall, is a 4 billion parameter Qwen-based text encoder built around the SMART DoRA training system for Z-Image Turbo. It is designed to transform minimal image prompts into detailed, high-fidelity visual narratives by adding explicit scene composition, lighting direction, material texture, and depth separation, while removing generic prompt elements. It also functions as a direct replacement for the stock Z-Image text encoder.

Key Capabilities

  • Prompt Enhancement: Rewrites simple concepts into descriptive, structured visual prompts locally.
  • Text Encoder Swap: Replaces the default Z-Image Qwen text encoder to alter image conditioning from the same seed.
  • Hybrid Mode: Can rewrite a prompt and then encode it, driving the image generation process.
  • Private Local Workflow: Optimized for use with LM Studio, ComfyUI, and llama.cpp for privacy and local execution.
  • Advanced Training: Utilizes SMART DoRA (Weight-Decomposed Low-Rank Adaptation) with entropic, holographic, topological, and manifold regularizers to prevent repetitive outputs and improve structural coherence.

Good for

  • Users seeking to generate more detailed and visually rich images from concise prompts.
  • Developers integrating a powerful, local prompt enhancement and text encoding solution into ComfyUI or LM Studio workflows.
  • Anyone looking to replace the default Z-Image text encoder for different creative outputs.
  • Workflows requiring a private, offline image prompt engineering solution without external API calls.

Finetunes
2 models