Date: 27.12.2025

Stable Diffusion 1.5 vs 2.0 vs XL Test

Test environment

• Workstation 40 GB RAM, 500GB SSD, 750W Power supply
• Ubuntu 24.04 LTS HWE Kernel
• Install python 3.12

My test environment: HP Z440 + NVIDIA RTX 3090

Steps

Preapre python environment for CUDA:

mkdir -p ~/llm && cd ~/llm
python3 -m venv .venv_llm_sd1.5
source ./.venv_llm_sd1.5/bin/activate
python -m pip install --upgrade pip
pip install "torch==2.7.0" "torchvision==0.22.0" "torchaudio==2.7.0" --index-url https://download.pytorch.org/whl/cu128
pip install transformers accelerate diffusers safetensors

Get the the StableDiffusion

• Get SD1.5

git lfs install
git clone https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5 sd1.5

• Get SD2.0

git lfs install
git clone https://huggingface.co/sd2-community/stable-diffusion-2 sd2.0

• Get SDXL

git lfs install
git clone https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0 sdxl

Create script test_cuda_sd1.5.py:

from diffusers import StableDiffusionPipeline
import torch

print("GPU available:", torch.cuda.is_available())
print("GPU name:", torch.cuda.get_device_name(0))

model_path = "/home/sysadmin/llm/sd1.5"

pipe = StableDiffusionPipeline.from_pretrained(
    model_path,
    torch_dtype=torch.float16,
    safety_checker=None,
    feature_extractor=None,
    use_safetensors=True,
    local_files_only=True
).to("cuda")

seed = torch.seed() # 4619748823278046644
generator = torch.Generator("cuda").manual_seed(seed)

prompt = "3d render of a cute stylized cat sitting on a chair, \
big expressive eyes, soft studio lighting, \
smooth shading, high detail, depth of field"
negative_prompt = "lowres, blurry, uncanny, deformed, \
bad anatomy, extra limbs, extra legs, multiple cats, \
cropped, watermark, text, logo, grain, noise, jpeg artifacts"

out = pipe(
    prompt=prompt,
    negative_prompt=negative_prompt,
    height=512, width=512,
    guidance_scale=9,
    clip_skip=1,
    num_inference_steps=50, 
    generator=generator)
image = out.images[0]

image.save(f"test_image_sd15_{seed}.png", format="PNG")

• Run test

python test_cuda_sd1.5.py

Create script test_cuda_sd2.0.py:

from diffusers import StableDiffusionPipeline
import torch

print("GPU available:", torch.cuda.is_available())
print("GPU name:", torch.cuda.get_device_name(0))

model_path = "/home/sysadmin/llm/sd2.0"

pipe = StableDiffusionPipeline.from_pretrained(
    model_path,
    torch_dtype=torch.float16,
    safety_checker=None,
    feature_extractor=None,
    use_safetensors=True,
    local_files_only=True
).to("cuda")

seed = torch.seed() # 885271215860566002
generator = torch.Generator("cuda").manual_seed(seed)

prompt = "3d render of a cute stylized cat sitting \
on a chair, big expressive eyes, soft studio lighting, \
smooth shading, high detail, depth of field"
negative_prompt = "lowres, blurry, uncanny, deformed, \
bad anatomy, extra limbs, extra legs, multiple cats, \
cropped, watermark, text, logo, grain, noise, jpeg artifacts"

out = pipe(
    prompt=prompt,
    negative_prompt=negative_prompt,
    height=768, width=768, 
    guidance_scale=8,
    clip_skip=1,
    num_inference_steps=45, 
    generator=generator)
image = out.images[0]

image.save(f"test_image_sd20_{seed}.png", format="PNG")

• Run test

python test_cuda_sd2.0.py

Create script test_cuda_sdxl.py:

from diffusers import StableDiffusionXLPipeline
import torch

print("GPU available:", torch.cuda.is_available())
print("GPU name:", torch.cuda.get_device_name(0))

model_path = "/home/sysadmin/llm/sdxl"

pipe = StableDiffusionXLPipeline.from_pretrained(
    model_path,
    torch_dtype=torch.float16,
    safety_checker=None,
    feature_extractor=None,
    local_files_only=True
).to("cuda")

seed = torch.seed()
generator = torch.Generator("cuda").manual_seed(seed)

prompt = "3d render, cute stylized cat sitting on \
a chair, big expressive eyes, soft studio lighting, \
smooth shading, high detail, clean background, \
cinematic, depth of field"
negative_prompt = "lowres, blurry, uncanny, deformed, \
bad anatomy, extra limbs, extra legs, multiple cats, \
cropped, watermark, text, logo, realistic fur photo, \
grain, noise"

out = pipe(
    prompt=prompt,
    negative_prompt=negative_prompt,
    height=1024, width=1024, 
    guidance_scale=5,
    clip_skip=1,
    num_inference_steps=45, 
    generator=generator)
image = out.images[0]

image.save(f"test_image_sdxl_{seed}.png", format="PNG")

• Run test

python test_cuda_sdxl.py

Enjoy the result!