US_Vintage_Poster

US Vintage Poster

A vintage poster style LoRA inspired by early-to-mid 20th century American posters, including WPA artwork, travel advertisements, National Park posters, aviation posters, wartime posters, and classic commercial lithographic designs.

This LoRA recreates the visual language of classic American printed posters:

• vintage travel advertisement illustration
• WPA Federal Art Project aesthetics
• National Park and tourism posters
• aviation and adventure posters
• wartime and industrial poster design
• classic commercial advertising illustration
• hand-painted lithograph style
• screen printing texture
• old paper appearance
• bold graphic composition
• limited color palette

The goal is not historical reconstruction, but a creative exploration of the visual language of 20th century American poster design.

---

✨ Features

• 1930s–1950s American vintage poster style
• WPA-inspired graphic design
• Classic travel poster aesthetics
• National Park promotional artwork
• Aviation and adventure poster composition
• Wartime and industrial poster visual language
• Commercial advertising illustration style
• Hand-painted lithograph feeling
• Screen print texture
• Retro typography layout
• Aged paper appearance
• Art Deco and mid-century graphic design influence

---

🖥️ Training Details

Hardware

Training was performed on:

• GPU: NVIDIA RTX PRO 6000 Blackwell 96GB

Configuration

• Base Model: Qwen/Qwen-Image
• LoRA Rank: 32
• Batch Size: 8
• Steps: 1750

---

⚠️ Important: Base Model Compatibility

This LoRA was trained and tested with:

Qwen/Qwen-Image

It was not trained or tested with Qwen/Qwen-Image-2512.

Compatibility with Qwen-Image-2512 or other Qwen-Image versions is not guaranteed. Using this LoRA with another base model may produce different results.

For the best experience, please use the original Qwen Image Text-to-Image model.

---

🚫 About NSFW

This LoRA was not trained for NSFW generation.

The purpose of this project is:

• vintage poster style
• historical-inspired artwork
• graphic design exploration
• creative image generation

If you are looking for NSFW-specific LoRAs, please explore other third-party projects.

---

🏞️ A Personal Note

This project was inspired by my personal experience exploring the American wilderness.

I have visited more than 60 National Parks across the United States, and every journey left a deep impression on me — the vast landscapes, the changing colors of the mountains, the silence of the deserts, and the feeling of standing in places shaped by millions of years of nature.

The classic American travel posters from the 1930s–1950s captured something unique:

• the spirit of exploration
• the romance of the open road
• the beauty of national landscapes
• the dream of discovering the unknown

Of course, an AI-generated image is not a historical reproduction.

When you ask:

"Generate a 1930s American National Park poster"

there is no single correct answer.

The result depends on:

• historical references
• artistic interpretation
• your prompt
• the model's imagination

This LoRA is my attempt to recreate that feeling — the nostalgia, adventure, and visual language of a time when posters inspired people to travel and discover the world.

Feel free to edit prompts, experiment with different themes, and create your own journeys.

After all, imagining the past is also a creative journey.

---

🖼️ Examples

The following images were generated with this LoRA.

The prompts below are the original prompts used for generation.

Feel free to copy, modify, and experiment with them.

---

Example 1

Prompt

zl_vintage_v1, a glamorous Hollywood actress in the 1930s, wearing an elegant evening gown, standing under movie theater lights, classic American movie poster, Golden Age Hollywood glamour, Art Deco design, vintage lithograph print, hand painted illustration, 1930s American advertising poster, aged paper texture, elegant feminine beauty, soft lighting.

---

Example 2

Prompt

zl_vintage_v1, 1930s American aviation poster, beautiful woman wearing pilot goggles and vintage flight suit, standing beside a classic airplane, adventurous spirit, Art Deco travel advertisement, hand painted poster art, vintage lithograph, 1930s airline advertisement, aged paper texture, heroic and elegant composition.

---

Example 3

Prompt

zl_vintage_v1, 1930s Grand Canyon National Park poster, dramatic canyon landscape at sunrise, layered red rock formations, Colorado River below, vast American wilderness, National Park Service travel postcard, WPA Federal Art Project style, hand painted lithograph, vintage tourism poster, aged paper grain.

---

🚀 Inference Example

• This example loads the full Qwen-Image model in BF16 without quantization or CPU offloading. A high-memory CUDA GPU is required.
• It uses Qwen-Image-Lightning 4-step acceleration for faster generation.
• An 80 GB GPU is recommended for reliable inference. GPUs with 48 GB of VRAM may work depending on resolution, software environment, and available system memory.

---

Installation

pip install -U diffusers transformers accelerate peft safetensors

---

Python

import math

import torch
from diffusers import (
    DiffusionPipeline,
    FlowMatchEulerDiscreteScheduler,
)

# =========================
# Model configuration
# =========================

BASE_MODEL_ID = "Qwen/Qwen-Image"

LIGHTNING_MODEL_ID = "lightx2v/Qwen-Image-Lightning"
LIGHTNING_WEIGHT_NAME = (
    "Qwen-Image-Lightning-4steps-V2.0-bf16.safetensors"
)

LORA_MODEL_ID = "perpetual3x/US_Vintage_Poster"
LORA_WEIGHT_NAME = "usvintage.safetensors"

DEVICE = "cuda"
DTYPE = torch.bfloat16

# =========================
# Scheduler
# =========================

scheduler_config = {
    "base_image_seq_len": 256,
    "base_shift": math.log(3),
    "invert_sigmas": False,
    "max_image_seq_len": 8192,
    "max_shift": math.log(3),
    "num_train_timesteps": 1000,
    "shift": 1.0,
    "shift_terminal": None,
    "stochastic_sampling": False,
    "time_shift_type": "exponential",
    "use_beta_sigmas": False,
    "use_dynamic_shifting": True,
    "use_exponential_sigmas": False,
    "use_karras_sigmas": False,
}

scheduler = FlowMatchEulerDiscreteScheduler.from_config(
    scheduler_config
)

# =========================
# Load Qwen-Image
# =========================

pipe = DiffusionPipeline.from_pretrained(
    BASE_MODEL_ID,
    scheduler=scheduler,
    torch_dtype=DTYPE,
).to(DEVICE)

# =========================
# Fuse 4-step Lightning LoRA
# =========================

pipe.load_lora_weights(
    LIGHTNING_MODEL_ID,
    weight_name=LIGHTNING_WEIGHT_NAME,
    adapter_name="lightning",
)

pipe.fuse_lora(
    adapter_names=["lightning"],
    lora_scale=1.0,
)

# Lightning has been fused into the base model.
pipe.unload_lora_weights()

# =========================
# Load US Vintage Poster LoRA
# =========================

pipe.load_lora_weights(
    LORA_MODEL_ID,
    weight_name=LORA_WEIGHT_NAME,
    adapter_name="vintage_style",
)

pipe.set_adapters(
    ["vintage_style"],
    adapter_weights=[1.0],
)

# =========================
# Generate
# =========================

prompt = (
    "zl_vintage_v1, 1930s American travel poster, "
    "Yellowstone National Park, Old Faithful geyser, "
    "WPA Federal Art Project style, vintage lithograph print"
)

negative_prompt = (
    "low quality, blurry, distorted, modern photography, "
    "photorealistic, watermark, signature"
)

generator = torch.Generator(
    device=DEVICE
).manual_seed(42)

image = pipe(
    prompt=prompt,
    negative_prompt=negative_prompt,
    width=1328,
    height=2048,
    num_inference_steps=4,
    true_cfg_scale=1.0,
    generator=generator,
).images[0]

image.save("us_vintage_poster.png")

---

License & Copyright Notice

This LoRA may be used for personal and non-commercial purposes only.

You may:

• Use this LoRA to create personal artwork
• Experiment with AI image generation
• Share generated images with proper attribution

You may not:

• Use this LoRA for commercial purposes
• Sell generated content as a commercial product
• Redistribute, repackage, or sell this LoRA
• Include this LoRA in paid services or products

For commercial licensing inquiries, please contact the author.