Skip to content
Local AI Wizard
Your machine
Theme

FLUX.2 klein 4B vs Z-Image Turbo for fast images

Two small, fast text-to-image models for a 12 to 16 GB card

Facts from the reviewed catalog

Facts from the reviewed catalog
SpecificationFLUX.2 [klein] 4BZ-Image Turbo
TaskImageImage
Parameters4B6B
LicenseApache 2.0Apache 2.0
Commercial useAllowedlicense textAllowedlicense text
Quality · LMArena 2026-10-07 · text-to-image1,030 ± 31,027–1,033 · 152,389 votes · rank 70 · source1,083 ± 51,078–1,089 · 24,808 votes · rank 59 · source
Memorypublisher's documented path from 13 GB (CUDA)publisher's documented path from 16 GB (CUDA)
Checkpointofficial cardofficial card

In LMArena 2026-10-07 · text-to-image, Z-Image Turbo rates 1,083 ± 5 against 1,030 ± 3 for FLUX.2 [klein] 4B, 53 points ahead with intervals that do not overlap. Blind pairwise votes by people on hosted outputs, CC BY 4.0. A rating is relative; overlapping intervals mean the order is not settled. Scale: Arena rating ± half the confidence interval.

Memory on three machines

Each machine's own reserve is added (graphics cards 2 GB; unified and system memory 25%, 2 to 8 GB). Planning estimates, not measured runs.

FLUX.2 klein 4B vs Z-Image Turbo for fast images: memory on three machines
MachineFLUX.2 [klein] 4BZ-Image Turbo
RTX 3060 · 12 GB VRAM13 GBBelow the documented minimum16 GBBelow the documented minimum
RTX 4060 Ti · 16 GB VRAM13 GBWithin a documented path16 GBWithin a documented path
RTX 4090 · 24 GB VRAM13 GBWithin a documented path16 GBWithin a documented path

FLUX.2 [klein] 4B: publisher's documented path from 13 GB (CUDA). Z-Image Turbo: publisher's documented path from 16 GB (CUDA). Formulas

Everything that runs on each machine: RTX 3060 · 12 GB VRAM · RTX 4060 Ti · 16 GB VRAM · RTX 4090 · 24 GB VRAM

Which to choose

Both are small, distilled text-to-image models under Apache 2.0, so commercial use has no extra conditions. FLUX.2 [klein] 4B has 4 billion parameters and Black Forest Labs documents a 13 GB CUDA path; Z-Image Turbo has 6 billion and Alibaba documents 16 GB. That puts FLUX.2 klein 4B within a 16 GB card's memory with some room, and Z-Image Turbo exactly at it. In the same LMArena text-to-image release Z-Image Turbo rates higher; the intervals are below.

Files and run commands

The default file of the most downloaded collected source, or the commands each card documents. Not run by this site.

FLUX.2 [klein] 4B

Documented ways to run it

Copied as printed in the linked source (read 2026-10-18); not run by this site.

Setup stated first by the source · Diffusers
pip install git+https://github.com/huggingface/diffusers.git
Diffusers · model card · **Using with Diffusers 🧨**
import torch
from diffusers import Flux2KleinPipeline

device = "cuda"
dtype = torch.bfloat16

pipe = Flux2KleinPipeline.from_pretrained("black-forest-labs/FLUX.2-klein-4B", torch_dtype=dtype)
pipe.enable_model_cpu_offload()  # save some VRAM by offloading the model to CPU

prompt = "A cat holding a sign that says hello world"
image = pipe(
    prompt=prompt,
    height=1024,
    width=1024,
    guidance_scale=1.0,
    num_inference_steps=4,
    generator=torch.Generator(device=device).manual_seed(0)
).images[0]
image.save("flux-klein.png")

Z-Image Turbo

Documented ways to run it

Copied as printed in the linked source (read 2026-10-18); not run by this site.

Setup stated first by the source · Diffusers
pip install git+https://github.com/huggingface/diffusers
Diffusers · model card · 🚀 Quick Start
import torch
from diffusers import ZImagePipeline

# 1. Load the pipeline
# Use bfloat16 for optimal performance on supported GPUs
pipe = ZImagePipeline.from_pretrained(
    "Tongyi-MAI/Z-Image-Turbo",
    torch_dtype=torch.bfloat16,
    low_cpu_mem_usage=False,
)
pipe.to("cuda")

# [Optional] Attention Backend
# Diffusers uses SDPA by default. Switch to Flash Attention for better efficiency if supported:
# pipe.transformer.set_attention_backend("flash")    # Enable Flash-Attention-2
# pipe.transformer.set_attention_backend("_flash_3") # Enable Flash-Attention-3

# [Optional] Model Compilation
# Compiling the DiT model accelerates inference, but the first run will take longer to compile.
# pipe.transformer.compile()

# [Optional] CPU Offloading
# Enable CPU offloading for memory-constrained devices.
# pipe.enable_model_cpu_offload()

prompt = "Young Chinese woman in red Hanfu, intricate embroidery. Impeccable makeup, red floral forehead pattern. Elaborate high bun, golden phoenix headdress, red flowers, beads. Holds round folding fan with lady, trees, bird. Neon lightning-bolt lamp (⚡️), bright yellow glow, above extended left palm. Soft-lit outdoor night background, silhouetted tiered pagoda (西安大雁塔), blurred colorful distant lights."

# 2. Generate Image
image = pipe(
    prompt=prompt,
    height=1024,
    width=1024,
    num_inference_steps=9,  # This actually results in 8 DiT forwards
    guidance_scale=0.0,     # Guidance should be 0 for the Turbo models
    generator=torch.Generator("cuda").manual_seed(42),
).images[0]

image.save("example.png")

Numbers each publisher reports

Each card's own results with its own settings. Publishers choose different tests and settings, so these values are never compared across publishers here.

FLUX.2 [klein] 4B · reported by Black Forest Labs

The publisher's card states no benchmark result for this checkpoint (read 2026-10-25). not published on the card · read 2026-10-25

Z-Image Turbo · reported by Alibaba

The publisher's card states no benchmark result for this checkpoint (read 2026-10-26). not published on the card · read 2026-10-26

Limits and dates

The published quality benchmark does not test a local quantized build. Memory estimates are not speed measurements or guaranteed hardware fits. Repository size, when available, is not VRAM. Formulas

Page updated · Catalog reviewed · LMArena 2026-10-07 · text-to-image