Z-Image (base) on 6 GB of VRAM
Not entirely. Even Q5_K_M is 1.6 GB over 6 GB, so ComfyUI keeps part of it in system RAM. The undistilled base of Z-Image Turbo: same 6B size, but 28–50 steps with CFG. Better for LoRA training and negative prompts.
Nothing fits entirely, but Q5_K_M (5.6 GB) overflows by only about 1.6 GB. ComfyUI keeps that part in system RAM automatically: slower than a full fit, but usable.
Also try the INT8 file (6.2 GB) on NVIDIA GPUs. It does not fit entirely, but a current ComfyUI streams the overflow from RAM and native 8-bit files often beat a small GGUF that does fit. Dynamic VRAM →
01Other amounts of VRAM
02Every file, on 6 GB
| File | Size | Needed | On this card | Quality | Download |
|---|---|---|---|---|---|
| 16-bit SAFETENSORS · Comfy-Org | 12.3 GB | 14.3 GB | 8.3 GB too big | the original weights | Hugging Face → |
| Q8_0 GGUF · unsloth | 7.2 GB | 9.2 GB | 3.2 GB too big | practically identical to the original | Hugging Face → |
| INT8 SAFETENSORS · Comfy-Org | 6.2 GB | 8.2 GB | 2.2 GB too big | practically identical to the original | Hugging Face → |
| Q6_K GGUF · unsloth | 6.1 GB | 8.1 GB | 2.1 GB too big | very close to the original | Hugging Face → |
| Q5_K_M ← GGUF · unsloth | 5.6 GB | 7.6 GB | spills 1.6 GB | close; small differences in fine detail | Hugging Face → |
| Q4_K_M GGUF · unsloth | 5.1 GB | 7.1 GB | spills 1.1 GB | good; some loss in fine detail and text | Hugging Face → |
| Q3_K_M GGUF · unsloth | 4.6 GB | 6.6 GB | spills 0.6 GB | noticeable loss of detail | Hugging Face → |
| Q2_K GGUF · unsloth | 4.0 GB | 6.0 GB | spills 0.0 GB | heavy loss; a last resort | Hugging Face → |
“Needed” adds 1.2 GB of working memory for Z-Image and a 0.8 GB reserve. Sizes 2026-10-01.
03What to download
| Part | File | Folder | Size | |
|---|---|---|---|---|
| Model | z-image-Q5_K_M.gguf Q5_K_M · unsloth/Z-Image-GGUF | models/unet | 5.6 GB | Download → |
| Text encoder | qwen_3_4b_fp8_mixed.safetensors Qwen3 4B FP8 · Comfy-Org/z_image | models/text_encoders | 5.6 GB | Download → |
| VAE | ae.safetensors FLUX.1 VAE (ae) · Comfy-Org/z_image | models/vae | 0.3 GB | Download → |
| Total download · keep about the same free on disk | 11.5 GB | |||
Alternatives: Qwen3 4B BF16 (8.0 GB). Sizes from Hugging Face, 2026-10-01.
Free ComfyUI workflow, tested on my RTX 5060 Ti: 16-bit workflow (.json). Drag the file into ComfyUI. All workflows →
04The text encoder
At 5.6 GB even the smallest Qwen3 4B file is too big for 6 GB, so it runs from system RAM. That slows down reading the prompt, not the picture. Sizes: 16-bit 8.0 GB, FP8 5.6 GB. Why the encoder does not count against VRAM →