🚀 Z-Image Turbo FP8 Hires Workflow (Low VRAM Optimized)
This is a high-efficiency ComfyUI workflow designed specifically for Low VRAM users. By utilizing FP8 Quantized Models and Latent Upscale technology, it generates high-resolution images (1024x1792) rapidly while maintaining minimal resource usage.
✨ Key Features
Extreme Low VRAM Usage: Full FP8 pipeline (Model & Text Encoder) to drastically reduce memory footprint.
Lightning Fast: Optimized for Turbo models and efficient sampling steps.
Hires Fix Pipeline: Utilizes
Latent Upscale+2nd Pass KSamplerto ensure crisp details without heavy VRAM cost.AuraFlow Architecture: Optimized using the
ModelSamplingAuraFlownode.
📂 Models Required & Downloads
To ensure the workflow functions correctly, please download the following models and place them in your respective ComfyUI folders:
1. UNet Model (Place in models/unet/)
File Name:
z-image-turbo-fp8-e4m3fn.safetensorsDownload: HuggingFace - Z-Image-Turbo-FP8
2. CLIP / Text Encoder (Place in models/clip/)
File Name:
qwen3-4b-fp8-scaled.safetensorsDownload: HuggingFace - Qwen3-4B-FP8
⚙️ Key Settings & Configuration
This workflow operates on a 2-Pass system. Please adhere to the following settings for the best results:
🔹 Phase 1: Base Generation
Latent Size: Generates at a lower initial resolution (e.g., 512x896) to save compute resources.
🔹 Phase 2: Latent Upscale
Upscale Method: Uses
LatentUpscaleBy.Scale Factor: Default is
2(resulting in a final output of 1024x1792).
🔹 Phase 3: Hires Fix (Refiner)
This step is crucial for image clarity and detail:
Sampler:
res_multistep(Highly Recommended).Denoise: Recommended range
0.5 - 0.6.< 0.5: Changes are minimal; the image may remain slightly blurry.> 0.6: Adds more detail, but setting this too high may alter the image structure or cause hallucinations.
📊 Performance Benchmark
Data based on actual testing:
GPUOutput ResolutionTimeNVIDIA RTX 5070 Ti1024 x 17928 ~ 9 sec
📝 Usage Tips
Memory Management: If you are extremely limited on VRAM, ensure no other large models are loaded in the background.
Prompting: Since this uses the Qwen text encoder, it has strong natural language understanding. Detailed, sentence-based prompts work very well.
Troubleshooting: If you notice the image details breaking or looking "burnt," try slightly lowering the
denoisevalue in the second KSampler.
Description
FAQ
Comments (9)
I have an RTX 3060 12GB graphics card. Is this workflow designed for my graphics card? My images generate in 39-50 seconds. Has anyone else with the same hardware had similar rendering times? Thanks for the workflow OP B)
Yes. I have the same GPU and it works a treat, about 20 seconds per image. The quality for a small model is superb.
@hoyleontour588 Okay, that's very interesting. I'll have to do some digging on my computer to optimize it a bit. Thanks for the reply <3
@hoyleontour588 Would you like to tell how much RAM you have? I have 16 GB and I suspect it's the bottleneck.
With this exact workflow and its default parameters (prompt, image size, steps, sampler, scheduler...) I get the image after 31-32 seconds.
I have 3060 12GB and 64GB RAM.
Oh, and I'm using the quantized GGUF model z_image_turbo-Q8_0.
@asmodaj Okay, thanks for the info. It looks like I need to rethink why my computer is so slow.
@Lombard99 It is probably the RAM, as you correctly said. If there is some offloading to the CPU involved, the amount of RAM could definitely be the bottleneck.
Thank you for this simple workflow!
Which nodes does it uses? I can't find most of them