Z-Image-Turbo/Base-AIO - Z-Image-Turbo-NVFP4-AIO

NSFW

🚀 Z-Image AIO Collection

⚡ Base & Turbo • All-in-One • Bilingual Text • Qwen3-4B

⚠️ IMPORTANT: Requires ComfyUI v0.11.0+

📥 Download ComfyUI

✨ What is Z-Image AIO?

Z-Image AIO is an All-in-One repackage of Alibaba Tongyi Lab's 6B parameter image generation models.

Everything integrated:

✅ VAE already built-in
✅ Qwen3-4B Text Encoder integrated
✅ Just download and generate!

🎯 Available Versions

🔥 Z-Image-Turbo-AIO (8 Steps • CFG 1.0)

Ultra-fast generation for production & daily use

⚫ NVFP4-AIO (7.8 GB) 🆕

🎯 ONLY for NVIDIA Blackwell GPUs (RTX 50xx)!
⚡ Maximum speed optimized
💾 Smallest file size
🚀 FP4 precision - blazing fast

Perfect for: RTX 5070, 5080, 5090 owners who want maximum speed

🟡 FP8-AIO (10 GB) ⭐ RECOMMENDED

✅ Best balance of size & quality
✅ Works on 8GB VRAM
✅ Fast downloads
✅ Ideal for most users

Perfect for: Daily use, testing, RTX 3060/4060/4070

🔵 FP16-AIO (20 GB)

💾 Same file size as BF16
🔄 ComfyUI auto-casts to BF16 for compute
⚠️ Does NOT enable FP16 compute mode
📦 Alternative download option

Note: Z-Image does not support FP16 compute - activation values exceed FP16's max range, causing NaN/black images. Weights are cast to BF16 during inference regardless of file format.

Perfect for: Alternative to BF16 download (identical inference behavior)

🌟 BF16-AIO (20 GB) ⭐ RECOMMENDED FOR FULL PRECISION

✅ BFloat16 full precision
✅ Absolute best quality
✅ Professional projects
✅ Also works on 8GB VRAM

Perfect for: Professional work, maximum quality

🎨 Z-Image-Base-AIO (28-50 Steps • CFG 3-5)

Full creative control for pros & LoRA training

🟡 FP8-AIO (10 GB)

✅ Efficient for daily use
✅ Full CFG control
✅ Negative prompts supported
✅ 8GB VRAM compatible

Perfect for: Daily work with full control

🔵 FP16-AIO (20 GB)

💾 Same file size as BF16
🔄 ComfyUI auto-casts to BF16 for compute
⚠️ Does NOT enable FP16 compute mode
📦 Alternative download option

Note: See technical explanation in FAQ below.

Perfect for: Alternative to BF16 download (identical inference behavior)

🌟 BF16-AIO (20 GB) ⭐ RECOMMENDED FOR FULL PRECISION

✅ Maximum quality
✅ Ideal for LoRA training
✅ Professional projects
✅ Highest precision

Perfect for: LoRA training, professional work

🆚 Turbo vs Base - When to Use?

⚡ Use TURBO when:

⚡ Speed is priority → 8 steps = 3-10 seconds
📸 Production workflows → Consistent high quality
💾 Quick iterations → Rapid prototyping
🎯 Simple prompts → Less complex scenes

🎨 Use BASE when:

🎨 Creative exploration → Higher diversity
🔧 LoRA/ControlNet dev → Undistilled foundation
📝 Complex prompting → Full CFG control
🚫 Negative prompts needed → Remove unwanted elements

⚙️ Recommended Settings

⚡ Turbo Settings (incl. NVFP4)

📊 Steps: 8
🎚️ CFG: 1.0 (don't change!)
🎲 Sampler: res_multistep OR euler_ancestral
📈 Scheduler: simple OR beta
📐 Resolution: 1920×1088 (recommended)
🚫 Negative Prompt: ❌ Not used!

🎨 Base Settings

📊 Steps: 28-50
🎚️ CFG: 3.0-5.0 (start with 4.0)
🎲 Sampler: euler ⭐ OR dpmpp_2m
📈 Scheduler: normal ⭐ OR karras
📐 Resolution: 512×512 to 2048×2048
🚫 Negative Prompt: ✅ Fully supported!

📊 Quick Overview

Turbo Versions

⚫ NVFP4  │ 7.8 GB  │ RTX 50xx only  │ Max Speed 🆕
🟡 FP8   │ 10 GB   │ 8GB VRAM       │ Recommended ⭐
🔵 FP16  │ 20 GB   │ → BF16 compute │ See FAQ ⚠️
🌟 BF16  │ 20 GB   │ 8GB VRAM       │ Max Quality ⭐

Base Versions

🟡 FP8   │ 10 GB   │ 8GB VRAM       │ Efficient
🔵 FP16  │ 20 GB   │ → BF16 compute │ See FAQ ⚠️
🌟 BF16  │ 20 GB   │ 8GB VRAM       │ LoRA Training ⭐

💡 Prompting Guide

✅ Good Example:

Professional food photography of artisan breakfast plate. 
Golden poached eggs on sourdough toast, crispy bacon, fresh 
avocado slices. Morning sunlight creating warm glow. Shallow 
depth of field, magazine-quality presentation.

❌ Bad Example:

breakfast, eggs, bacon, toast, food, morning, plate

📝 Tips

DO:

✅ Use natural language
✅ Be detailed (100-300 words)
✅ Describe lighting & mood
✅ Specify camera angle
✅ English OR Chinese (or both!)

DON'T:

❌ Tag-style prompts (tag1, tag2, tag3)
❌ Very short prompts (under 50 words)
❌ Negative prompts with Turbo

🌐 Bilingual Text Rendering

English:

Neon sign reading "OPEN 24/7" in bright blue letters 
above entrance. Modern sans-serif font, glowing effect.

中文:

Traditional tea house entrance with sign reading 
"古韵茶坊" in elegant gold Chinese calligraphy.

Both:

Modern cafe with bilingual sign. "Morning Brew" in 
white script above, "晨曦咖啡" in Chinese below.

📥 Installation

Step 1: Download

Choose your version based on:

GPU: RTX 50xx → NVFP4 possible
VRAM: 8GB → FP8 recommended
Purpose: LoRA Training → Base BF16

Step 2: Place File

ComfyUI/models/checkpoints/
└── Z-Image-Turbo-FP8-AIO.safetensors

Step 3: Load & Generate

Open ComfyUI (v0.11.0+!)
Use "Load Checkpoint" node
Select your AIO version
Generate!

No separate VAE or Text Encoder needed!

🙏 Credits

Original Model

👨‍💻 Developer: Tongyi Lab (Alibaba Group)
🏗️ Architecture: Single-Stream DiT (6B parameters)
📜 License: Apache 2.0

Links

🔗 Z-Image Base: https://huggingface.co/Tongyi-MAI/Z-Image

🔗 Z-Image Turbo: https://huggingface.co/Tongyi-MAI/Z-Image-Turbo

🧠 Text Encoder: https://huggingface.co/Qwen/Qwen3-4B

📈 Version History

v2.2 - FP16 Clarification

📝 Updated FP16 descriptions for technical accuracy
⚠️ Clarified: FP16 weights ≠ FP16 compute
🔄 FP16 files are cast to BF16 during inference

v2.1 - NVFP4 Release 🆕

➕ Z-Image-Turbo-NVFP4-AIO (7.8GB)
⚡ Optimized for NVIDIA Blackwell (RTX 50xx)
🚀 Maximum speed generation

v2.0 - Base AIO Release

➕ Z-Image-Base-BF16-AIO
➕ Z-Image-Base-FP16-AIO
➕ Z-Image-Base-FP8-AIO
🔄 ComfyUI v0.11.0+ support
📝 Qwen3-4B Text Encoder

v1.1 - FP16 Added

➕ Z-Image-Turbo-FP16-AIO
🔧 Wider GPU compatibility

v1.0 - Initial Release

✅ Z-Image-Turbo-FP8-AIO
✅ Z-Image-Turbo-BF16-AIO
✅ Integrated VAE + Text Encoder

❓ FAQ

Q: Which version should I choose?

RTX 50xx + Speed → NVFP4 🆕
Most users       → Turbo FP8 ⭐
Full precision   → BF16 ⭐
LoRA Training    → Base BF16

Q: Turbo or Base?

Fast & simple    → Turbo ⚡
Full control     → Base 🎨

Q: Will NVFP4 work on my RTX 4090?

❌ No! NVFP4 is only for RTX 50xx (Blackwell architecture).

Use FP8 instead for RTX 40xx and older.

Q: Do I need separate VAE/Text Encoder?

❌ No! Everything is already integrated.

Just Load Checkpoint and go!

Q: Works on 8GB VRAM?

✅ Yes! All versions work on 8GB VRAM.

(NVFP4 requires RTX 50xx regardless of VRAM)

⚠️ Q: What about FP16 for older GPUs (RTX 2000/3000)?

Important technical clarification:

Z-Image does NOT support FP16 compute type. Here's why:

📊 Technical reason:
- FP16 max value: ~65,504
- BF16 max value: ~3.39e+38 (same as FP32)
- Z-Image's activation values exceed FP16's range
- Result: Overflow → NaN → Black images

What actually happens:

ComfyUI automatically casts weights to BF16 for computation
You can see this in logs: "model weight dtype X, manual cast: torch.bfloat16"
"Weight dtype" (file format) ≠ "Compute dtype" (actual calculation)

For RTX 20xx users (no native BF16):

BF16 is emulated via FP32 = slower but works
There is no way to run Z-Image in true FP16 compute
FP8 with CPU offload may be a better option for limited VRAM

TL;DR: FP16 and BF16 files behave identically during inference. Choose based on download preference, not GPU compatibility.

🚀 Get Started Now!

Download → Load Checkpoint → Generate!

Recommended versions:

🟡 FP8 for most users (best size/quality balance)
🌟 BF16 for maximum quality
⚫ NVFP4 for RTX 50xx speed

All versions work on 8GB VRAM

Happy generating! 🎨

Description

Z-Image-Turbo-AIO-NVFP4

FAQ

Comments (23)

makoshark1975Feb 1, 2026· 3 reactions

CivitAI

Can you tell me which app you are using to merge things? I have two checkpoints I want to test by merging them and I am a bit new.

SeeSeeLP

Author

Feb 1, 2026

Hello, no apps scripts like this one, which I write myself:

def merge_flux2_aio(unet_path, te_path, vae_path, output_path, te_type="qwen3_4b"): """ Merge UNET, Text-Encoder, and VAE into a single AIO checkpoint. Expected output structure: - model.diffusion_model.* (UNET) - text_encoders.<te_type>.transformer.model.* (Text-Encoder model weights) - text_encoders.<te_type>.logit_scale (Text-Encoder logit scale) - vae.* (VAE) """ print("\n" + "="*60) print("Flux2 Klein AIO Merger") print("="*60) print("\n[1/4] Loading components...") unet_sd = load_safetensors(unet_path) te_sd = load_safetensors(te_path) vae_sd = load_safetensors(vae_path) print("\n[2/4] Processing keys...") merged_sd = {} unet_sample_key = list(unet_sd.keys())[0] if unet_sd else "" if unet_sample_key.startswith("model.diffusion_model."): print(" UNET: Already has 'model.diffusion_model.' prefix") for k, v in unet_sd.items(): merged_sd[k] = v elif unet_sample_key.startswith("diffusion_model."): print(" UNET: Adding 'model.' prefix") for k, v in unet_sd.items(): merged_sd[f"model.{k}"] = v else: print(" UNET: Adding 'model.diffusion_model.' prefix") for k, v in unet_sd.items(): merged_sd[f"model.diffusion_model.{k}"] = v te_sample_key = list(te_sd.keys())[0] if te_sd else "" te_prefix = f"text_encoders.{te_type}." if te_sample_key.startswith("text_encoders."): print(f" Text-Encoder: Already has 'text_encoders.' prefix") for k, v in te_sd.items(): merged_sd[k] = v elif te_sample_key.startswith("model."): print(f" Text-Encoder: Adding '{te_prefix}transformer.' prefix") for k, v in te_sd.items(): merged_sd[f"{te_prefix}transformer.{k}"] = v merged_sd[f"{te_prefix}logit_scale"] = torch.tensor(4.6055) else: print(f" Text-Encoder: Adding '{te_prefix}transformer.model.' prefix (unknown input format)") for k, v in te_sd.items(): merged_sd[f"{te_prefix}transformer.model.{k}"] = v merged_sd[f"{te_prefix}logit_scale"] = torch.tensor(4.6055) vae_sample_key = list(vae_sd.keys())[0] if vae_sd else "" if vae_sample_key.startswith("vae."): print(" VAE: Already has 'vae.' prefix") for k, v in vae_sd.items(): merged_sd[k] = v else: print(" VAE: Adding 'vae.' prefix") for k, v in vae_sd.items(): merged_sd[f"vae.{k}"] = v print("\n[3/4] Merged checkpoint summary:") get_key_prefix_info(merged_sd, "AIO") unet_keys = len([k for k in merged_sd if k.startswith("model.")]) te_keys = len([k for k in merged_sd if k.startswith("text_encoders.")]) vae_keys = len([k for k in merged_sd if k.startswith("vae.")]) print(f" - UNET keys: {unet_keys}") print(f" - Text-Encoder keys: {te_keys}") print(f" - VAE keys: {vae_keys}") print(f" - Total keys: {len(merged_sd)}") print(f"\n[4/4] Saving to: {output_path}") os.makedirs(os.path.dirname(output_path) if os.path.dirname(output_path) else ".", exist_ok=True) save_file(merged_sd, output_path) file_size = os.path.getsize(output_path) / (1024 * 1024 * 1024) print(f" File size: {file_size:.2f} GB") print("\n" + "="*60) print("Merge complete!") print("="*60)

SeeSeeLP

Author

Feb 1, 2026

It looks good in the preview ^^ So, if you can't program it yourself, you can probably find something online; maybe there are even programs that can do it. I've heard there are also nodes for ComfyUI.

makoshark1975Feb 1, 2026

@SeeSeeLP Yeah prolly a little beyond me at the moment, I just didn't know if koyha SS or something could merge checkpoints

SeeSeeLP

Author

Feb 1, 2026

@makoshark1975 yes, as I said, there should be nodes for this in comfyui, I just googled it

makoshark1975Feb 1, 2026

@SeeSeeLP I don't use comfyui, was looking for an alternative to it

rivdemon1221554Feb 1, 2026· 10 reactions

CivitAI

The idea of z-image base is great, but without fine-tunes and some kind of pre-fill element, it's garbage. It expects far too much written detail, which is of course of good thing for people who just rip prompts using other VL systems, but why should someone need to do that? Remember SDXL, where you could just say 'a cat in a cowboy suit', and it would generate something amazing? Now you need to write a damn paragraph to get something here. Definitely not going anywhere in the mainstream sense, that's for sure. Anyway, still love your work as always, especially the Qwen one.

SeeSeeLP

Author

Feb 1, 2026

Thanks for your feedback.

Something about Qwen will be coming soon, as soon as I finally get away from LTX2. ^^ I'm currently playing around with parameters to ensure that the image doesn't shift to a different look during an image-to-video workflow.

Astolfo2001Feb 4, 2026· 1 reaction

CivitAI

GGUF version when?

DieselMar 1, 2026· 1 reaction

CivitAI

There is no link to the Z-Image-Turbo-NVFP4-AIO,
base (zImageTurboBaseAIO_zImageTurboNVFP4AIO) is downloaded instead.

SeeSeeLP

Author

Mar 2, 2026· 1 reaction

That's the correct version, the name is just awful ^^ Next time I'll make sure that such a name mess doesn't happen. But thanks for your comment.

Wendy_EarthMar 21, 2026· 1 reaction

CivitAI

I don't know why i am not able to load this model on 8gbvram before it was working but now it keeps getting crash @SeeSeeLP can you plz guide me on this(I am using your workflow).

SeeSeeLP

Author

Mar 21, 2026

@Wendy_Earth

Sorry to hear that. If it was working before on 8GB VRAM, then it’s probably not your GPU itself, but something that changed in ComfyUI, a node update, or the model settings.

Please check if ComfyUI or any custom nodes were updated recently. A different VAE, resolution, batch size, or loader setting can also cause higher VRAM usage and crashes.

Send me the error message of the crash log and I’ll try to help.

Wendy_EarthMar 21, 2026

@SeeSeeLP It keep getting silent crash on load checkpoint

Wendy_EarthMar 21, 2026

I have RTX 4060 8Gb VRAM and 16 GB RAM

Wendy_EarthMar 21, 2026

@SeeSeeLP I didn't updated ComfyUI and I am using the same workflow which you provided so the version is same rightnow i have comfyui version v0.5.1

SeeSeeLP

Author

Mar 21, 2026

@Wendy_Earth

ComfyUI 0.18.0

ComfyUI_frontend v1.41.21

Templates v0.9.26

LoRA Manager v1.0.0-stable

EasyUse v1.3.7

ComfyUI-Manager V3.39.2

rgthree-comfy v1.0.2512112053

Wendy_EarthMar 21, 2026

@SeeSeeLP worked Thank you so much.

blhllMar 22, 2026· 1 reaction

CivitAI

Moodie's WF: ZIB+ZIT (using your's AIO base +AIO Turbo) = the most realistic images with the most realistic skin without any loras. That is if you aim for realism. AIO versions on their own dont even come close.

SeeSeeLP

Author

Mar 22, 2026

Thanks for the tip 👌

CornmeisterNLMar 25, 2026

@blhll got a link to Moodies WF ?

blhllMar 25, 2026

@CornmeisterNL yeah, here it is: https://civitai.com/models/2345999?modelVersionId=2786477

CornmeisterNLMar 27, 2026

@blhll tnx!!

Checkpoint

ZImageTurbo