SoLordZ ZIT/ZIB 8steps + SoLordZ TE — GGUF FULL Bundle
Quantized GGUF versions of SoLordZ ZIT/ZIB 8steps (Z-Image Turbo) and the companion SoLordZ Text Encoder (uncensored Qwen3-4B fine-tune).
Original model downloaded from Civitai: SoLordZ ZIT/ZIB/Krea2 8steps.
Original text encoder downloaded from Civitai: THM_UNC_Z-image_TE.
Converted for use with ComfyUI and with stable-diffusion.cpp directly.
⚠️ Important — How this model works
This model runs in "turbo mode": 8 steps, CFG=1.
It ignores negative prompts. Everything you want to exclude must be written directly in the positive prompt — especially for anatomy, where control is crucial.
Recommended sampler settings from the author:
- First pass:
er_sde/beta - Second pass:
euler/simple - For realism:
euler/normal - For art: sometimes better with
euler_a/normal
Why this bundle
The base Z-Image model (up to 50 steps, CFG 5-9) is flexible but slow. SoLordZ's goal was ambitious: keep the quality of the "slow" model, but teach it to generate in 8 steps at CFG=1.
Key improvements over the base model:
- Increased detail in complex scenes
- Reduced background blurring (less blur, more coherent depth)
- Improved realism across all styles
- Better prompt adherence — the model understands complex, descriptive prompts more accurately
This bundle pairs the model with its official companion text encoder — an uncensored fine-tune of Qwen3-4B that was specifically trained for Z-Image. Using them together gives the best results, as confirmed by the authors and community feedback.
📦 What's in this repository
Diffusion model (root folder)
| File | Size | Notes |
|---|---|---|
zit_SoLordZ_v2-Q3_K.gguf | 4.31 GB | For very tight VRAM budgets |
zit_SoLordZ_v2-Q4_K.gguf | 5.15 GB | Recommended for 8 GB VRAM |
zit_SoLordZ_v2-Q4_K_S.gguf | 4.83 GB | Slightly smaller than Q4_K |
zit_SoLordZ_v2-Q5_K.gguf | 5.68 GB | Better detail, still fits 8 GB |
zit_SoLordZ_v2-Q6_K.gguf | 6.05 GB | For 12+ GB cards |
zit_SoLordZ_v2-Q8_0.gguf | 7.34 GB | Near-lossless, needs 12+ GB |
Text encoder (clip/ folder)
| File | Size | Notes |
|---|---|---|
Qwen3-4b-SoLordZ-Q4_K_M.gguf | ~2.4 GB | Recommended for 8 GB VRAM |
Qwen3-4b-SoLordZ-Q5_K_M.gguf | 2.89 GB | Good balance |
Qwen3-4b-SoLordZ-Q6_K.gguf | 3.31 GB | For 12+ GB cards |
Qwen3-4b-SoLordZ-Q8_0.gguf | 4.28 GB | Maximum quality |
Qwen3-4B-Q4_K_M.gguf | 2.50 GB | Standard Qwen3-4B (alternative) |
VAE (vae/ folder)
| File | Size | Notes |
|---|---|---|
ae.safetensors | 335 MB | 16-channel Z-Image VAE |
Recommended combo for GTX 1070 (8 GB): diffusion Q4_K + text encoder Q4_K_M + VAE.
🚀 ComfyUI workflow
The working path is martin-rizzo/ComfyUI-ZImagePowerNodes. It loads Z-Image GGUF directly and applies the correct latent/conditioning shapes.
Required nodes
cd ComfyUI/custom_nodes
git clone https://github.com/martin-rizzo/ComfyUI-ZImagePowerNodes.git
Alternative: standard ComfyUI GGUF nodes If you prefer the built-in UnetLoader (GGUF), you need city96/ComfyUI-GGUF on PR #392 or later:
cd ComfyUI/custom_nodes/ComfyUI-GGUF
git fetch origin pull/392/head:pr392
git checkout pr392
That fixes the x_pad_token shape mismatch. Power Nodes is more reliable for Z-Image, though.
🙏 Credits Original model: SoLordZ ZIT/ZIB/Krea2 8steps by SoLordZ and Triple_Headed_Monkey on Civitai.
Original text encoder: THM_UNC_Z-image_TE by Triple_Headed_Monkey and SoLordZ on Civitai.
Base architecture: Z-Image Turbo by Tongyi Lab (Alibaba).
ComfyUI nodes: martin-rizzo/ComfyUI-ZImagePowerNodes.
Conversion toolchain: city96/ComfyUI-GGUF + llama.cpp.
Runtime: leejet/stable-diffusion.cpp.
Quantization and packaging: gimly13.
📄 License All components in this repository are distributed under CreativeML OpenRAIL-M:
Original diffusion checkpoint (SoLordZ ZIT/ZIB) — CreativeML OpenRAIL-M.
SoLordZ text encoder (uncensored Qwen3-4B fine-tune) — CreativeML OpenRAIL-M.
Standard Qwen3-4B — Apache 2.0 (Alibaba).
Z-Image VAE (ae.safetensors) — Apache 2.0 (Tongyi Lab).
See the original Civitai pages for full licensing details.