Z-Image Turbo is a distilled version of Z-Image, a 6B image model based on the Lumina architecture, developed by the Tongyi Lab team at Alibaba Group. Source: https://huggingface.co/Tongyi-MAI/Z-Image-Turbo

I've uploaded quantized versions from bf16 to fp8, meaning the weights had their precision - and consequently their size - halved for a substantial performance boost while keeping most of the quality. Inference time should be similar to regular "undistilled" SDXL, with better prompt adherence and resolution/details. Ideal for weak(er) PCs.

Features

• Lightweight: the Turbo version was trained at low steps (5-15), and the fp8 quantization is roughly 6 GB in size, making it accessible even to low-end GPUs.

• Uncensored: many concepts censored by other models (<cough> Flux <cough>) are doable out of the box.

• Good prompt adherence: comparable to Flux.1 Dev's, thanks to its powerful text encoder Qwen 3 4B.

• Text rendering: comparable to Flux.1 Dev's, some say it's even better despite being 2x smaller (probably not as good as Qwen Image's though).

• Style flexibility: capable of generating photorealistic images, as well as anime, painting, pixel art, low poly, comics, illustration, pop art, etc.

• High resolution: capable of generating up to 2MP resolution natively (before upscaling!).

Dependencies

• Download Qwen 3 4B to your text_encoders directory: https://huggingface.co/Comfy-Org/z_image_turbo/blob/main/split_files/text_encoders/qwen_3_4b.safetensors

• Download Flux VAE to your vae directory: https://huggingface.co/Comfy-Org/z_image_turbo/blob/main/split_files/vae/ae.safetensors

Instructions

Workflow and metadata are available in the showcase images.

• Steps: 5 - 15.

• CFG: 1.0. This will ignore negative prompts, so no need for them.

• Sampler/scheduler: depends on the art style. Here are my findings so far:

  • Photorealistic:

    • Favourite combination for the base image: euler + beta, simple or bong_tangent (from RES4LYF) - fast and good even at low (5) steps.

    • Most multistep samplers (e.g.: res_2s, res_2m, dpmpp_2m_sde etc) are great, but some will be 40% slower at same steps. They might require a scheduler like sgm_uniform.

    • Almost any sampler will work fine - sa_solver, seeds_2, er_sde, gradient_estimation.

    • What you probably want to avoid (at least in the base image) due to bad results or poor performance:

      • dpm_adaptive sampler

      • karras scheduler

    • Some samplers and schedulers add too much texture, you can adjust it by increasing the shift (e.g.: set shift 7 in ComfyUI's ModelSamplingAuraFlow node).

  • Illustrations (e.g.: anime):

    • res_2m or rk_beta produce sharper and more colourful results.

  • Others:

    • I'm still testing. Use euler + simple just to be safe for now.

• Resolution: up to 2MP native. When in doubt, use same as SDXL, Flux.1, Qwen Image, etc (it works even as low as 512px like SD 1.5 times). Some examples:

  • 896x1152

  • 1024x1024

  • 1216x832

  • 1440x1440

  • 1024x1536

• Upscale and/or detailers are recommended to fix smaller details like eyes, teeth, hair. See my workflow embedded in the main cover image.

  • Instead of using a sampler with low step like we do with other models, Z-Image works best when you use something like KSampler (Advanced) (in ComfyUI), or any node that allows you to set a starting step.

  • set shift to 7 while using something like euler + simple (some samplers/schedulers might have their own shift, which won't help), this will prevent exaggerated textures and noise.

• Prompting: officially they say long and detailed prompts in natural language works best, but I tested with comma-separated keywords/tags, JSON, whatever... either should work fine. Keep it in English or Mandarin for more accurate results.

FAQ

• Is the model uncensored?

  • Yes, it might just not be well trained on the specific concept you're after. Try it yourself.

• Why do I get too much texture after upscaling?

  • See instructions about upscaling above.

• Does it run on my PC?

  • If you can run SDXL, chances are you can run Z-Image Turbo fp8. If not, might be a good time to purchase more RAM or VRAM.

  • All my images were generated on a laptop with 32GB RAM, RTX3080 Mobile 8GB VRAM.

• Is the license permissive?

  • It's Apache 2.0, so quite permissive.