⚡ Z-Image Turbo Premium TXT-2-IMG Workflow ⚡

➡️ Grab your copy of the v5 Ultimate Edition now 🔥

WF Features:

✨ Compact Pro GUI Design || SDXL Pre-Definer (NSFW Fix) || 6x Detailer || SeedVR2 Upscaler || Seed Variance Enhancer || Lora Loader || Color Grading || 5 different v5 versions 👀

Z-Image Turbo is a fast, powerful and highly efficient image generation model with 6B parameters. The strength of this model lies in its ability to generate photorealistic images. But even though the model follows prompts very well, the output still isn’t premium in terms of quality. We still need an upscaling process here to make the images presentable. This is where this workflow comes in! Including the ADetailer, SDXL Pre-Definer as NSFW Fix and SeedVR2 Upscaler for high quality outputs!

Version differences:

• v1 = Base Workflow with Ultimate SD Upscaler (outdated)

• v2 = Lora Loader + Ultimate SD Upscaler (outdated)

• v3 = LL + Color Grading + 2x Detailer + Ultimate SD Upscaler (Great quality)

• v4 = LL + Color Grading + 2x Detailer + SeedVR2 Upscaler (Best quality)

• v5 = LL + Color Grading + SDXL PreDefiner + 6x Detailer + SeedVR2 Upscaler (Best 4 NSFW)

Recommendations based on your skill level:

• ComfyUI Beginner 👉🏼 v5 Basic Version ⚡

• Advanced User 👉🏼 v4 Version or v5 Pro ⚡⚡

• AI Pro Creator 👉🏼 v5 Ultimate Edition ⚡⚡⚡

I have completely revised the manual so that even ComfyUI beginners can set up and use the Ultimate Edition! 🫶🏼

🚀 What do you need to get started right away?

• This Workflow - just download it, unzip the file and load the *.json into ComfyUI

• Z-Image Model: z_image_turbo_bf16.safetensors (place in \models\diffusion_models\)

• Text encoder file: qwen_3_4b.safetensors (place in \models\text_encoders\)

• The VAE file: ae.safetensors (place in \models\vae\)

• Face Detector: face_yolov8m.pt (place in \models\ultralytics\bbox\)

• Eyes Detector: Eyeful_v2-Paired.pt (place in \models\ultralytics\bbox\)

• Hand Detector: hand_yolov8s.pt (place in \models\ultralytics\bbox\)

• Various upscale models of your choice for v1, v2 & v3 (place in \models\upscale_models\)

Regarding the upscale model (only important for v1, v2 & v3) I recommend 4x_NMKD-Siax_200k.pth or 4x_UltraSharp.pth. These deliver the best results. But feel free to try out different upscale models. You can find a good selection here. You can ignore this if you use the SeedVR2 Version (v4+). I recommend using the SeedVR2 Version because it is faster and brings better image quality.

For SeedVR2 Upscaler you can find the model files here. Place one or more of them in \models\SEEDVR2\ folder. For more detailed info's about the models scroll down to the v4 Changelog.

If you used FLUX before then you already have the ae.safetensors file and dont need to download it again. But if you prefer more details you can try out the UltraFlux-v1 VAE file.

You can also use the fp8 checkpoint if you run a gpu with 8gb of vram.

For v5 Versions you also need some good SDXL (or Pony / Illustrious) models. Place them in \models\checkpoints\

🚨 Make sure you use the latest stable ComfyUI Core version. Also don't forget to update the ComfyUI Frontend package manually and your Custom Nodes via ComfyUI Manager. Otherwise compatibility problems may occur and the workflow is not working properly.

The v3, v4 and v5 versions are made with the following packages:

• ComfyUI Core v0.3.75

• ComfyUI Frontend v1.34.2

• ComfyUI Manager v3.37.2

• Python Version 3.12.10

• PyTorch Version 2.8.0+cu128

• All latest versions of the custom nodes listed below

Please ensure that you use at least these versions. Otherwise, I cannot provide support if something is not working for you.

❓ Do I need some special nodes to run this workflow?

The nodes inside these workflows are common nodes. If ComfyUI still indicates a missing node just install it via ComfyUI Manager and restart briefly.

Custom nodes used in this workflow (v5):

• rgthree-comfy

• ComfyUI-Impact-Pack

• ComfyUI-Easy-Use

• ComfyUI_UltimateSDUpscale

• ComfyUI-Art-Venture

• ComfyUI-Impact-Subpack

• ComfyUI-Moonpack

• ControlAltAI_Nodes

• VRGameDevGirl Video Enhancement Nodes

• ComfyUI-SeedVR2_VideoUpscaler

• ComfyUI-Flux-Continuum

• WAS Node Suite (Revised)

• SeedVarianceEnhancer

💡 Usage Tips:

• Simply set the desired resolution on the left, put in your prompt and start generating.

• Why there is no negative prompt field? If the CFG value is set to 1.0 the negative prompt is ignored. This is one of the reasons why the model is so fast. This is intentional and not a mistake. Other workflows are ignoring this fact and offer a negative input node. But you don't need a negative prompt here. Simply enter your positive prompt and you're good to go!

• For prompting, it's recommended to use natural language in complete sentences. Describe the person or people, the setting, the camera angle, and the lighting. The more detailed you are, the more accurate the result. 100-300 words works best.

• If you only want to generate raw images for testing purposes, simply toggle off the upscaling and detailer groups using the first node on the top-left. This is also recommended for color grading. Don't forget to fix the seed for that.

• Feel free to experiment with the denoise values ​​in the detailer nodes to get the perfect result. Also try out different upscale models.

• You don't need to go higher than 9 steps with this model. It is a 8 Steps Model. It will not generate better images if you go higher. Just leave it on 8 or max 9 steps.

• I like to have the option to directly select the aspect ratio (e.g. 5:7 or 9:16) in the node. If you prefer to enter the resolution manually simply disconnect or delete the "Base Resolution Node". Afterwards you can enter your desired resolution in the "latent image" node.

✅ v2 Changelog:

In comparison to my other results with SDXL I wasn't entirely satisfied. Therefore, I revised and optimized the v1 workflow:

• ✨ Added Toggle On/Off Node for Upscaling Group

Toggling the entire upscaler group just to generate RAW images was cumbersome. Now, a single on/off switch at the start lets you enable or disable the upscaler with just one click – quick and easy.

• ✨ Added Lora Loader Node

With more Lora's being released for Z-Image Turbo these days, this node lets you load and configure multiple Loras at once. Making experimentation faster and easier.

• ✨ Added Clear VRAM and Clear Cache Node

Placed between the Base Image Generation and Upscaler sections, these nodes improve memory management. Especially on GPUs with limited VRAM and slightly speed up the workflow.

• ✨ Added Seed Generator Node

The new Seed Generator node is connected to the KSampler and also to the Upscaler node. This means that the Upscaler receives the same seed. I feel this leads to better results.

• ✨ Added fast sharpen / unsharpen node

This node allows you to adjust the sharpness slightly after upscaling. Feel free to experiment with thestrengthvalue. However, I suggest leaving it at 0.50.

• ✨ Added fast film grain node

This node adds film grain after upscaling. This provides a professional finish and a cinematic look. The grain_intensity value can be increased up to 0.050. I find less is more here and therefore have it set to 0.010.

• ✨ Adjusted some parameters for better performance

I adjusted some parameters, mainly affecting the Image Upscale node. I found that 20 steps unnecessarily slowed down the upscaling process and resulted in a poorer outcome, especially regarding skin texture. Eight steps proved to be the perfect value. This reduced the overall workflow time to 30% while simultaneously improving the quality.

🛠️ All the changes combined now result in a faster workflow and very decent images compared to v1.

✅ v3 Changelog:

The results with version 2 of the workflow are already quite good. But there's always room for improvement. So I've made some further adjustments that will help you get better images. What's changed?

• ✨ Added Color Grading Node

Since the release of the Z-Image model, generating thousands of images showed that it struggles with following light and color grading prompts. Images often look overbrighten and like they were taken with a high-end smartphone, making it hard to achieve a cinematic look. To fix this, a color grading node has been added to the first block (next to the LoRa loader). Here, you can adjust temperature, brightness, contrast, saturation, and gamma to get truly impressive results. For best results, set the seed in the "Seed Generator" node to "fixed," disable the detailer and upscaler, and experiment a bit. This lets you achieve a cinematic look immediately - no LoRa needed.

• ✨ Added Adetailer Group for Image Improvement

To complement the upscaling process, a dedicated face and eye detailer section has been added. Bare feet and hands already render well in 99% of cases with Z-Image, so no detailer is needed for them. Adding one would only slow down the workflow unnecessarily. To keep the overall workflow compact despite the new nodes, the relevant parts have been moved into subgraphs. You still have full control over key parameters such as steps, CFG, and denoise for both detailers. Too small changes indicate a denoise value that’s too low, while issues with eyes or other image artifacts mean the denoise value is too high. Feel free to experiment with the denoise to get the best results.

Make sure that you have downloaded the 2 needed files for the detailer:

• Face Detector: face_yolov8m.pt (place in \models\ultralytics\bbox)

• Eyes Detector: Eyeful_v2-Paired.pt (place in \models\ultralytics\bbox)

Without these files the workflow will just stop at the detailer group!

• ✨ Added Adetailer Group to the Toggle On / Off Node

The new detailer section has been integrated into the Toggle On/Off node at the start of the workflow. This allows you to enable or disable it with a single click. It’s especially useful when generating RAW images, testing color grading, or looking for a good seed without applying additional detail enhancements.

• ✨ Added Calculated Resolution Node

To make it easy to see which base resolution is being used, I've added a node that displays this information. It's the result of the selected resolution in megapixels and the aspect ratio. You can see this "Calculated Resolution" node between the "Base Resolution" and "Latent Image" nodes on the left.

• ✨ Added Model Sampling Aura Flow Node

In other Z-Image workflows, you'll see the "Model Sampling Aura Flow" node, but it is usually muted. Since I didn’t find it useful for generating more variations from the same prompt, I removed it in versions 1 and 2 of the workflow. I’ve now added it back so you can test it yourself. This doesn’t affect runtime, and of course, the node can still be muted if desired.

• ✨ Adjusted some parameters

I’ve tweaked a few settings again to achieve the best possible results. First, I increased the base resolution from 1.2MP to 2.0MP. This alone slightly improves image quality, as there are more pixels in the RAW image. If you have a powerful GPU, you can raise this value to 2.2MP or even 2.5MP. Additionally, I increased the sharpen strength from 0.50 to 0.55 and adjusted the film grain intensity from 0.010 to 0.015, which seems to be the sweet spot. Of course, feel free to experiment yourself.

🛠️ The download includes two versions: a full version with all updates, including the Detailer section, and a Lite version without the Detailer Group but with all other changes and color grading. Enjoy!

✅ v4 Changelog:

Even better? Yes, much better! 🔥

• ✨ Removed Ultimate SD Upscaler and replaced it with SeedVR2

At the request of some users, I have replaced the upscaler part (Ultimate SD Upscaler) with SeedVR2. This results in even better images and faster upscaling in v4 of this workflow!

The hole process (loading model, generating raw image, detailer part and upscaling with seedvr2) takes about 70-80 seconds on my RTX 5070TI. That's really great for a high res 4k output picture!

• ❓ Do you need to download any models for SeedVR2?

Yes, you can but you don't need to do. Just select your preferred model (e.g. seedvr2_ema_7b-Q4_K_M.gguf) on the 🗂️ SeedVR2 Load DiT Model Node and it will be automatically downloaded on the first run. Because of this the first run will take a little longer. You can open the console and check the download progress there.

• ❓ Which upscaling model should you use?

This really depends on the GPU and VRAM you are using. Just try out some of them. If you get a OOM-Error (Out of Memory) you know this model is too big in file size or your output resolution is too high. Then you need to use a smaller one.

A small overview:

• seedvr2_ema_7b-Q4_K_M.gguf (4.76GB)

• seedvr2_ema_7b-Q5_K_M.gguf (5.78GB)

• seedvr2_ema_7b-Q6_K.gguf (6.86GB)

• seedvr2_ema_7b-Q8_0.gguf (8.84GB)

⚠️ It is recommended to avoid models with “sharp” in their name (e.g. seedvr2_ema_7b_sharp-Q4_K_M.gguf), as they are known to introduce a dark, fine mesh or grid-like overlay into the generated images. This artifact is often only noticeable when zooming in, but can still negatively affect overall image quality.

⬇️ You can download these models from Huggingface and put them in the\models\SEEDVR2\ folder. The VAE file for SeedVr2 (ema_vae_fp16.safetensors) is also downloaded automatically.

Do not change the value on the Downscale 4 better Upscale Node because we need to downscale the image before we can upscale it. SeedVR2 is made for upscaling very small images. So this step is necessary. You choose your desired output resolution on the ✨ SeedVR2 Upscaler Node. There you can change the resolution value. You don't need to change any other settings on these SeedVR2 Nodes.

I've included two versions of the workflow in the v4 zip file. The first file is the updated workflow with SeedVR2 upscaler. However, I've also included a version where I've simply appended the SeedVR2 upscaler to the end. This means it first upscales with Ultimate SD Upscaler and then again with SeedVR2. This allows you to see a direct comparison and convince yourself that SeedVR2 is the better choice.

Enjoy upscaling with SeedVR2! 🚀

🆕 v5 Changelog:

Even better than the previous version? Yes! It's the same great output quality than v4 (no more improvements needed for quality) but with more options for specific needs. 😉

What's New?

• ✨ Removed Seed Generator Node

This node was only relevant for workflow versions v1 through v3. Now you can change or lock the seed directly in the KSampler node itself. Way cleaner and more straightforward.

• ✨ Added Seed Variance Enhancer Node

To get way more variety in your generations without constantly tweaking the prompt, I've thrown in a brand-new "Seed Variance Enhancer" node. This bad boy gives you that sweet SDXL-style variation we all love, just by tweaking a couple sliders. That said, handle with care! Crank the values too high and you'll end up with mutated bodies real quick. My default presets are pretty stable and safe, but feel free to play around with "randomize_percent" and "strength". Just don't go over 50 on randomize_percent or 30 on strength, or things can get ugly fast. Since it can be a bit wild, the node is disabled by Default. You can toggle it on/off with a single click on the far left whenever you want.

• ✨ Added Hand Detailer

Yeah, even Z-Image isn't flawless and loves spitting out hands with 4 or 6 fingers from time to time. So I decided to level up the Face and Eye Detailers by adding a dedicated Hand Detailer. It does a solid job fixing those wonky Hands. Not quite 100% success rate, but way better than before. To use this new node, just grab the hand_yolov8s.pt file and drop it into your \models\ultralytics\bbox\ folder. Super straightforward and you'll notice the difference in your generations right away.

• ✨ Added Detailer Steps Slider

To make tweaking the steps for all Detailer nodes super quick and painless, I've added a global Steps Slider node. It affects every Detailer at once. No more endless clicking through each one individually. Just slide it and watch the magic happen.

• ✨ Switched Pre-Definer to SDXL Model

This is the biggest and most important change in v5. Instead of sticking with the same Z-Image model for the Pre-Definer (which honestly didn't shake things up much), you can now select any SDXL model here! Yes, that includes Pony and Illustrious checkpoints too! This single switch fixes a ton of Z-Image's classic headaches right out of the gate. No more endless same-old faces. SDXL models bring way more diversity and freshness. Even better, they tackle those typical NSFW issues head-on: overly massive boobs, unrealistic nipples, and mutated or censored pussies are a thing of the past. Now you get the best of both worlds!

You've got a built-in NSFW fix now, no extra bullshit required. Say goodbye to those ZIT LoRA's that never really helped and always messed with your main image too much. On top of that, the SDXL Pre-Definer injects way more realistic skin details and textures into your generations.

Why does this work so well? Simple: SDXL models actually know what human anatomy is supposed to look like. From now on, just load up your favorite SDXL model and finally get clean, error-free NSFW ZIT pics without the frustration. I highly recommend testing a bunch of different checkpoints. Results vary wildly depending on the model and that's part of the fun. If the Pre-Definer feels like it's overpowering your raw image, just dial down the Denoise and CFG values a bit.

Models I personally recommend for this:

• EpicRealism XL

• CyberRealistic XL

• Juggernaut XL

• CyberRealistic Pony

or any other SDXL, Pony or Illustrious model that catches your eye. 👀

Since we're working with SDXL models for the pre-definer, we need separate prompt fields for that. You'll find my preset in the negative prompt. This works very well for most images. For the positive prompt, it's best to simply copy and paste the prompt from the Z-Image field.

• ✨ Added LoRa Loader for SDXL Pre-Definer

To let the SDXL Pre-Definer unleash its full potential, I've equipped it with a dedicated LoRa Loader. Got favorite SDXL LoRa's that aren't available for Z-Image yet? No problem anymore! You can now easily drop them right into this workflow. The LoRa effect applies only to the pre-defined Image. It won't bleed into the detailers or mess with your final output. Clean, targeted and powerful.

• ✨ Four Different v5 Versions

I've built four distinct v5 versions in total, each tailored for different use cases and skill levels.

1. Basic Version: ⚡

This one comes without the SDXL Pre-Definer. It generates the ZIT image straight away, runs it through the three core Detailers (Face, Eyes, Hands) and finishes with a clean SeedVR2 upscale. Perfect for anyone wanting fast, high-quality SFW images without any extra complexity. It's also the best starting point for ComfyUI newcomers. Simple, reliable and stunning results right out of the box.

2. Pro Versions: ⚡⚡

The download includes two flavors of the Pro version, both equipped with the powerful SDXL Pre-Definer we talked about earlier. The difference? One sticks with the Z-Image model for the Detailers, while the other runs all three Detailers through the SDXL model. The outputs are noticeably different. Just fire them up and test for yourself. Both have their strengths depending on what you're going for. Overall, the Pro version is great for NSFW generations thanks to that SDXL Pre-Definer magic!

3. Ultimate Edition: ⚡⚡⚡🔥

The Ultimate Edition of the v5 workflow is specially tuned and optimized for NSFW creations. Instead of just three standard Detailers, it packs six of them. Added dedicated ones for feet, breasts and pussies. The results? Absolutely fantastic, next-level detail that blows the Pro out of the water.

And here's the real game-changer: All six Detailers come with their own individual LoRa Loaders. That means you can slap specialized SDXL based LoRa's on every single detail for insane new possibilities. On top of that, each Detailer gets its own positive and negative prompt fields for even finer control over the final output.

The SDXL Pre-Definer shines even brighter here too. It's paired with an extra upscaler node that feeds in more pixels, giving the Pre-Definer and all six Detailers way more room to work their magic. You'll see the difference immediately.

This version is for power users who want maximum control over every little detail or anyone going commercial. The quality jump over the Pro is massive! 🔥

The Ultimate Edition isn't included in the free download here. You'll find a preview there though.

☕ Grab your full copy on KoFi ⬇️

It comes with 2 different versions of the workflow, a detailed guide and all six custom Ultralytic files. FREE Updates included!

Not a fan of Ko-Fi? You can also buy it on Patreon or Gumroad.

Or just hit me up via DM on CivitAI or Discord (@mrsmithfpv).

I still tried to keep every version as compact as possible with the same eye candy design you like. 👌

With v5 of this workflow I'm finally getting the ZIT NSFW images I've always wanted. Clean, realistic and hot as hell. I hope you do too! Have a blast with it!

Merry Christmas! 🎅🎄🎁

🆕 v5.1 Ultimate Edition Changelog:

I’ve been busy in the forge to bring you even more power and flexibility. This update adds a complete new version to the download: A dedicated Z-Image branch that swaps out the standard SDXL Detailers.

What’s new?

• ✨ Z-Image Power: You can now use specific ZIT LoRa's for every single detailer, pushing the final output quality further. It's like v4 on steroids! 👀

• ✨ Total Control: Added individual On/Off toggles for all 6 LoRa Loaders of the Detailers and the SDXL Pre-Definer. Just one click to customize your render flow. 🖱️

• ✨ GUI Update: I’ve decluttered the GUI by hiding the 'Clean Cache' and 'Clean VRAM' nodes. Now, only the nodes you actually need are visible, making the workflow much cleaner and easier to navigate. Less mess, more focus! 🧹

• ✨ The Master Guide: I’ve completely rewritten and massively expanded the Manual. It’s now your ultimate roadmap for perfect generations. 📖

Every supporter (and of course new supporter) can grab the updated files right now! The download includes both different versions of the Ultimate v5.1 and the updated manual.

Enjoy! And a happy new year! 🫶🏼

If you have ideas for optimization / new feature requests or just wanna say "thanks" feel free to leave a post in the comment section. If you have questions about this workflow, feel free to reach out to me on Discord anytime!

💕 Don't forget to spread some love - hit the like button and follow me (also on Ko-Fi and Patreon) to not miss any new updates and more workflows like this one! 👍

If you like these workflows you can also buy me a coffee! ☕

Have fun and feel free to post your generated images in the gallery below! I would like to see what you do with these workflows!😊

🥰 v5 Basic Version without Pre-Definer (incl. 3x ZIT Detailer):

🤩 v5 Pro Version with SDXL Pre-Definer (incl. 3x SDXL Detailer or 3x ZIT Detailer):

😎 Boss mode activated! v5 Ultimate Edition with better Pre-Definer & 6x SDXL Detailer or 6x ZIT Detailer (for Face, Eyes, Hands, Foot, Pussy & Nipples) + Bonus Features: