FLUX.2 dev vs Qwen-Image workflows
Choosing an image editing or text-to-image workflow
Facts from the reviewed catalog
| Specification | FLUX.2 [dev] | Qwen-Image |
|---|---|---|
| Task | Image | Image |
| Parameters | 32B | 27.1B |
| License | Publisher terms: FLUX [dev] Non-Commercial License | Apache 2.0 |
| Commercial use | Not allowedlicense text | Allowedlicense text |
| Quality · LMArena 2026-10-07 · text-to-image | 1,146 ± 41,142–1,150 · 77,988 votes · rank 38 · source | 1,057 ± 31,054–1,059 · 84,550 votes · rank 66 · source |
| Memory | publisher's documented path from 24 GB (CUDA, conditional) | publisher's documented path from 4 GB (CUDA, conditional) |
| Checkpoint | official card | official card |
In LMArena 2026-10-07 · text-to-image, FLUX.2 [dev] rates 1,146 ± 4 against 1,057 ± 3 for Qwen-Image, 89 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.
| Machine | FLUX.2 [dev] | Qwen-Image |
|---|---|---|
| RTX 4060 Ti · 16 GB VRAM | 24 GBBelow the documented minimum | 4 GBWithin a conditional path (offload or quantization) |
| RTX 4090 · 24 GB VRAM | 24 GBWithin a conditional path (offload or quantization) | 4 GBWithin a conditional path (offload or quantization) |
| RTX 5090 · 32 GB VRAM | 24 GBWithin a conditional path (offload or quantization) | 4 GBWithin a conditional path (offload or quantization) |
FLUX.2 [dev]: publisher's documented path from 24 GB (CUDA, conditional). Qwen-Image: publisher's documented path from 4 GB (CUDA, conditional). Formulas
Everything that runs on each machine: RTX 4060 Ti · 16 GB VRAM · RTX 4090 · 24 GB VRAM · RTX 5090 · 32 GB VRAM
Which to choose
The licence decides most cases. FLUX.2 [dev] ships under a non-commercial licence, so a business cannot use its outputs or weights commercially without another agreement; Qwen-Image is Apache 2.0. FLUX.2 [dev] is the larger model (32 billion parameters against 27.1) and is oriented to editing an existing image; Qwen-Image is a text-to-image model with editing variants. Both have a rating on the same LMArena text-to-image release, shown below with its interval.
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 [dev]
Documented ways to run it
Copied as printed in the linked source (read 2026-10-18); not run by this site.
ComfyUI · ComfyUI official examples · Basic Example Workflow: ComfyUI workflow
Qwen-Image
Documented ways to run it
Copied as printed in the linked source (read 2026-10-18); not run by this site.
pip install git+https://github.com/huggingface/diffusersfrom diffusers import DiffusionPipeline
import torch
model_name = "Qwen/Qwen-Image"
# Load the pipeline
if torch.cuda.is_available():
torch_dtype = torch.bfloat16
device = "cuda"
else:
torch_dtype = torch.float32
device = "cpu"
pipe = DiffusionPipeline.from_pretrained(model_name, torch_dtype=torch_dtype)
pipe = pipe.to(device)
positive_magic = {
"en": ", Ultra HD, 4K, cinematic composition.", # for english prompt
"zh": ", 超清,4K,电影级构图." # for chinese prompt
}
# Generate image
prompt = '''A coffee shop entrance features a chalkboard sign reading "Qwen Coffee 😊 $2 per cup," with a neon light beside it displaying "通义千问". Next to it hangs a poster showing a beautiful Chinese woman, and beneath the poster is written "π≈3.1415926-53589793-23846264-33832795-02384197". Ultra HD, 4K, cinematic composition'''
negative_prompt = " " # using an empty string if you do not have specific concept to remove
# Generate with different aspect ratios
aspect_ratios = {
"1:1": (1328, 1328),
"16:9": (1664, 928),
"9:16": (928, 1664),
"4:3": (1472, 1140),
"3:4": (1140, 1472),
"3:2": (1584, 1056),
"2:3": (1056, 1584),
}
width, height = aspect_ratios["16:9"]
image = pipe(
prompt=prompt + positive_magic["en"],
negative_prompt=negative_prompt,
width=width,
height=height,
num_inference_steps=50,
true_cfg_scale=4.0,
generator=torch.Generator(device="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 [dev] · reported by Black Forest Labs
Qwen-Image · reported by Qwen
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