Z-Image Turbo PiD Direct 4K Fast Image Generation Workflow

Watch the full video first if you want to understand how this Z-Image Turbo + PiD workflow works in practice. The video shows how a low-step Z-Image Turbo base generation can be pushed into a PiD 2K-to-4K enhancement lane, how the native result compares with the enhanced output, and how to launch the workflow online without building a local ComfyUI setup.

This ComfyUI workflow is designed for Z-Image Turbo direct 4K image generation using PiD as the high-resolution enhancement stage. Its main purpose is to combine the speed advantage of Z-Image Turbo with the high-resolution detail recovery of PiD. Instead of generating a slow high-resolution image from scratch, the workflow first creates a 1024×1024 base image through Z-Image Turbo, captures a PiD-ready latent, and then sends it into a 2K-to-4K PiD refinement pass.

The workflow is built around z_image_turbo_bf16.safetensors as the main UNet model. It uses qwen_3_4b.safetensors as the text encoder, the lumina2 CLIP type, ae.safetensors as the VAE, and EmptySD3LatentImage for the 1024×1024 base latent. The prompt is handled through PiDTextPrompt, which sends the same text into the positive prompt encoder and also provides the caption for PiD preparation. This keeps the base generation and the PiD enhancement stage aligned around the same visual instruction.

The core Z-Image Turbo generation lane uses ModelSamplingAuraFlow with shift 3, CLIPTextEncode for positive and negative conditioning, and PiDKSamplerCapture for both native latent generation and PiD latent capture. The sampling setup is optimized for speed: 9 steps, CFG 1, Euler sampler, beta scheduler, denoise 1, and capture_step 8. This matches the low-step Turbo logic and makes the workflow much faster than a heavier 25-step or 50-step image generation route.

After the native base latent is produced, the workflow decodes it through VAEDecodeTiled and saves it as the native baseline output. At the same time, PiDKSamplerCapture sends the captured PiD latent and sigma into PiDPrepare. PiDPrepare is configured with the zimage backbone, 2kto4k checkpoint type, scale 4, auto download enabled, and cleanup after prepare enabled. PiDSample then runs the heavy PiD pass with 4 PiD steps, CFG scale 1, fixed seed, aggressive cleanup, and sequential_blocks_aggressive offload. PiDFinalize converts the PiD sample into the final enhanced image.

Compared with ordinary Z-Image Turbo generation, this workflow does more than just create a fast image. It uses PiD to rebuild high-resolution structure after the Turbo base is already formed. Compared with a traditional external upscaler, this route is more integrated because it works from the captured latent and the original caption, not only from a finished flat image. This makes it useful for fantasy illustration, book-cover style images, cinematic posters, concept art, premium social media covers, RunningHub showcases, and Civitai workflow publishing.

Main features:

• Z-Image Turbo PiD direct 4K workflow

• Z-Image Turbo bf16 UNet route

• Qwen 3 4B text encoder

• AE VAE decoding

• 1024×1024 base latent generation

• PiDTextPrompt unified prompt and caption input

• ModelSamplingAuraFlow with shift 3

• Fast 9-step Turbo sampling setup

• CFG 1 low-step generation logic

• Euler / beta sampling route

• Capture step 8 for PiD preparation

• PiDPrepare with zimage backbone

• PiD 2K-to-4K enhancement route

• PiDSample with 4 PiD steps

• Sequential block offload and aggressive cleanup

• Native baseline and enhanced output comparison

Suggested workflow:

Start with a clear prompt that defines the subject, atmosphere, style, lighting, composition, and final texture target. Keep the base canvas at 1024×1024 for the first test. Run the native Z-Image Turbo lane first and check whether the composition is correct. If the base image is weak, adjust the prompt before entering PiD. Once the native result is stable, let PiDKSamplerCapture pass the captured latent and sigma into PiDPrepare, then run PiDSample and compare the native baseline with the enhanced output. If the final result becomes too heavy or unstable, keep PiD steps at 4 and rely on cleanup / offload settings instead of forcing more aggressive enhancement.

⚙️ RunningHub Workflow

Try the workflow online right now — no installation required.
👉 Workflow: https://www.runninghub.ai/post/2059961793564471298?inviteCode=rh-v1111

If the results meet your expectations, you can later deploy it locally for customization.

🎁 Fan Benefits: Register to get 1000 points + daily login 100 points — enjoy 4090 performance and 48 GB super power!

📺 Bilibili Updates (Mainland China & Asia-Pacific)

If you’re in the Asia-Pacific region, you can watch the video below to see the workflow demonstration and creative breakdown.
📺 Bilibili Video: https://www.bilibili.com/video/BV1tsVH6kEBJ/

☕ Support Me on Ko-fi

If you find my content helpful and want to support future creations, you can buy me a coffee ☕.
Every bit of support helps me keep creating — just like a spark that can ignite a blazing flame.
👉 Ko-fi: https://ko-fi.com/aiksk

💼 Business Contact

For collaboration or inquiries, please contact aiksk95 on WeChat.

⚙️打开下方链接即可在线体验，无需安装。
👉 工作流： https://www.runninghub.ai/post/2059961793564471298?inviteCode=rh-v1111
如果觉得效果理想，你也可以在本地进行自定义部署。

🎁 粉丝福利： 注册即送 1000 积分，每日登录 100 积分，畅玩 4090 体验 48 G 超级性能！

📺 Bilibili 更新（中国大陆及南亚太地区）

如果你在中国大陆或南亚太地区，可以通过下方视频查看该工作流的实测效果与构思讲解。
📺 B站视频： https://www.bilibili.com/video/BV1tsVH6kEBJ/

我会在 夸克网盘 持续更新模型资源：
👉 https://pan.quark.cn/s/20c6f6f8d87b
这些资源主要面向本地用户，方便进行创作与学习。