Skip to content
Local AI Wizard
Your machine
Theme

HunyuanImage 3.0 vs Qwen-Image 2512 for images

A very large and a mid-size open text-to-image model

Facts from the reviewed catalog

Facts from the reviewed catalog
SpecificationHunyuanImage 3.0 (Fal)Qwen-Image-2512
TaskImageImage
Parameters80B20B
LicensePublisher terms: Tencent Hunyuan Community License AgreementApache 2.0
Commercial useAllowed with conditionsNot licensed in the EU, UK or South Korea; over 100M monthly users needs a Tencent license · license textAllowedlicense text
Quality · LMArena 2026-10-07 · text-to-image1,150 ± 31,146–1,153 · 180,518 votes · rank 35 · source1,125 ± 31,122–1,129 · 98,288 votes · rank 44 · source
Memorypublisher's documented path from 240 GB (CUDA)labelled 16-bit estimate (~50 GB), no documented minimum
Checkpointofficial cardofficial card

In LMArena 2026-10-07 · text-to-image, HunyuanImage 3.0 (Fal) rates 1,150 ± 3 against 1,125 ± 3 for Qwen-Image-2512, 24 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.

HunyuanImage 3.0 vs Qwen-Image 2512 for images: memory on three machines
MachineHunyuanImage 3.0 (Fal)Qwen-Image-2512
RTX 5090 · 32 GB VRAM240 GBBelow the documented minimum~50 GBLess memory than the 16-bit estimate
DGX Spark · 128 GB unified240 GBBelow the documented minimum~50 GBEnough memory for the 16-bit estimate
Mac Studio M5 Max · 128 GB unifiedNo documented path for this platform (a CUDA figure is not carried over to a Mac)~50 GBEnough memory for the 16-bit estimate

HunyuanImage 3.0 (Fal): publisher's documented path from 240 GB (CUDA). Qwen-Image-2512: labelled 16-bit estimate (~50 GB), no documented minimum. Formulas

Everything that runs on each machine: RTX 5090 · 32 GB VRAM · DGX Spark · 128 GB unified · Mac Studio M5 Max · 128 GB unified

Which to choose

Size decides where they run. HunyuanImage 3.0 has 80 billion parameters and Tencent documents a CUDA path that starts at 240 GB of GPU memory, so no single desktop in the catalog runs it as published. Qwen-Image 2512 has 20 billion and is Apache 2.0; with no documented minimum, its rows use a labelled 16-bit estimate. Tencent's licence excludes the EU, the UK and South Korea. Both have a rating in the same LMArena text-to-image release, and their intervals are shown 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.

HunyuanImage 3.0 (Fal)

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 · Transformers
# Download from HuggingFace
hf download tencent/HunyuanImage-3.0 --local-dir ./HunyuanImage-3
Transformers · model card · 2️⃣ Run with Transformers
from transformers import AutoModelForCausalLM

# Load the model
model_id = "./HunyuanImage-3"
# Currently we can not load the model using HF model_id `tencent/HunyuanImage-3.0` directly 
# due to the dot in the name.

kwargs = dict(
    attn_implementation="sdpa",     # Use "flash_attention_2" if FlashAttention is installed
    trust_remote_code=True,
    torch_dtype="auto",
    device_map="auto",
    moe_impl="eager",   # Use "flashinfer" if FlashInfer is installed
)

model = AutoModelForCausalLM.from_pretrained(model_id, **kwargs)
model.load_tokenizer(model_id)

# generate the image
prompt = "A brown and white dog is running on the grass"
image = model.generate_image(prompt=prompt, stream=True)
image.save("image.png")
Setup stated first by the source · HunyuanImage
git clone https://github.com/Tencent-Hunyuan/HunyuanImage-3.0.git
cd HunyuanImage-3.0/
HunyuanImage · model card · 3️⃣ Run the Demo
# Without PE
export MODEL_PATH="./HunyuanImage-3"
python3 run_image_gen.py \
    --model-id $MODEL_PATH \
    --verbose 1 \
    --prompt "A brown and white dog is running on the grass" \
    --bot-task image \
    --image-size "1024x1024" \
    --save ./image.png \
    --moe-impl flashinfer

# With PE
export DEEPSEEK_KEY_ID="your_deepseek_key_id"
export DEEPSEEK_KEY_SECRET="your_deepseek_key_secret"
export MODEL_PATH="./HunyuanImage-3"
python3 run_image_gen.py \
    --model-id $MODEL_PATH \
    --verbose 1 \
    --prompt "A brown and white dog is running on the grass" \
    --bot-task image \
    --image-size "1024x1024" \
    --save ./image.png \
    --moe-impl flashinfer \
    --rewrite 1

Qwen-Image-2512

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
from diffusers import DiffusionPipeline
import torch

model_name = "Qwen/Qwen-Image-2512"

# 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).to(device)

# Generate image
prompt = '''A 20-year-old East Asian girl with delicate, charming features and large, bright brown eyes—expressive and lively, with a cheerful or subtly smiling expression. Her naturally wavy long hair is either loose or tied in twin ponytails. She has fair skin and light makeup accentuating her youthful freshness. She wears a modern, cute dress or relaxed outfit in bright, soft colors—lightweight fabric, minimalist cut. She stands indoors at an anime convention, surrounded by banners, posters, or stalls. Lighting is typical indoor illumination—no staged lighting—and the image resembles a casual iPhone snapshot: unpretentious composition, yet brimming with vivid, fresh, youthful charm.'''

negative_prompt = "低分辨率,低画质,肢体畸形,手指畸形,画面过饱和,蜡像感,人脸无细节,过度光滑,画面具有AI感。构图混乱。文字模糊,扭曲。"


# Generate with different aspect ratios
aspect_ratios = {
    "1:1": (1328, 1328),
    "16:9": (1664, 928),
    "9:16": (928, 1664),
    "4:3": (1472, 1104),
    "3:4": (1104, 1472),
    "3:2": (1584, 1056),
    "2:3": (1056, 1584),
}

width, height = aspect_ratios["16:9"]

image = pipe(
    prompt=prompt,
    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.

HunyuanImage 3.0 (Fal) · reported by Tencent

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

Qwen-Image-2512 · reported by Qwen

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

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