README.md

---
tags:
- text-to-image
- stable-diffusion
- lora
- diffusers
- template:sd-lora
widget:
- text: >-
    NASA_CONCEPT artist depiction of a space colony. Mid-century modern
    buildings under a glass dome, lush plants and waterfalls. Monorail seen in
    the background of the image. sci-fi
  output:
    url: images/5Hk8Kpjf1ZepeYTLPsD3S_1976870e9edb4212b6b6150e43c64d83.jpg
- text: >-
    NASA_CONCEPT Lunar surface Soviet base, 1960s concept art. Dome-shaped
    habitats connected by pressurized tunnels. Cosmonaut in foreground planting
    Soviet flag.
  output:
    url: images/p9NUH0PSGipjf34WR3ihf_976eb62ebb694002a4a953b27e593afd.jpg
- text: >-
    NASA_CONCEPT Rotating wheel space station, Soviet 1960s visualization. Large
    circular structure with multiple spokes leading to central hub. Visible
    living quarters along the wheel's rim. Docked Vostok-style spacecraft.
    Cosmonauts exercising in zero-G gym area. Communal dining hall with
    propaganda posters barely visible. External view showing solar arrays and
    communication equipment. Style mimics technical illustrations with cutaway
    sections revealing interior details. Cool blues and greys dominate, with
    occasional warm tones for human elements. Pioneering yet pragmatic
    atmosphere.
  output:
    url: images/GJP_i_P0PesdNwjQIOjnH_d4e5b9ffc185488486ae1a905cb1ef95.jpg
- text: >-
    NASA_CONCEPT Detailed cross-section of a space station module, 1960s style
    technical drawing. Cutaway view revealing multiple decks. Labeled components
    including life support systems, communication equipment, and scientific
    instruments. Color-coded sections for different functional areas. Inset
    diagrams showing specific systems like air recycling or water purification.
  output:
    url: images/kjskndOkIl_cbhQiiLtGo_517a7b546529496b8f4fa4ad6b340ee9.jpg
- text: NASA_CONCEPT Landmark 81, Ho Chi Minh City in 2030s
  output:
    url: images/example_dfflhfnh4.png
base_model: black-forest-labs/FLUX.1-dev
instance_prompt: NASA_CONCEPT
datasets:
- davanstrien/nasa_concept_art

---

# NASA Concept Art LoRA

This model is a LoRA (Low-Rank Adaptation) fine-tuning of black-forest-labs/FLUX.1-dev trained on a curated dataset of NASA concept art from the 1960s and 1970s. It specializes in generating retro-futuristic space-themed images inspired by NASA's historical vision of space exploration.

## Examples

<Gallery />

## Model Details

- **Base Model:** black-forest-labs/FLUX.1-dev
- **Training Data:** [davanstrien/nasa_concept_art](https://huggingface.co/datasets/davanstrien/nasa_concept_art)
- **LoRA Rank:** 16
- **Training Steps:** 1000
- **Learning Rate:** 0.0004
- **Batch Size:** 1
- **Caption Dropout Rate:** 0.05
- **Optimizer:** AdamW8bit

## Intended Use

This model is designed for creating artistic renditions of retro-futuristic space concepts. It's particularly suited for:

- Generating concept art for sci-fi projects set in a retro-futuristic universe
- Creating visual aids for educational content about the history of space exploration
- Artistic projects exploring the aesthetic of mid-20th century space age optimism

## Limitations

- The model's output is stylistically limited to the aesthetic of 1960s-1970s NASA concept art
- It may not accurately represent current or future space technology
- The small size of the training dataset (36 images) may limit the model's variability

## Ethical Considerations

Users should be aware that the generated images reflect historical concepts and may not represent diverse perspectives. Care should be taken not to present these images as accurate depictions of current or future NASA projects.

## How to Use

To get the best results, use the trigger word `NASA_CONCEPT` at the beginning of your prompt. For example:
```
NASA_CONCEPT artist depiction of a space colony. Mid-century modern buildings under a glass dome, lush plants and waterfalls. Monorail seen in the background of the image. sci-fi
```



## Training Details

The model was trained for 1000 steps using the AdamW8bit optimizer. The learning rate was set to 0.0004 with a batch size of 1. A caption dropout rate of 0.05 was used during training to improve generalization.

## License

This model uses images from NASA, which are generally not subject to copyright in the United States. However, users should refer to the [NASA_IMAGE_USAGE_LICENSE.txt](link_to_license_file) for full terms of use.

## Acknowledgments

- Original artwork created by various artists commissioned by NASA

For more information about the dataset, please refer to the [dataset card](https://huggingface.co/datasets/davanstrien/nasa_concept_art).

## Download Model

Weights for this model are available in Safetensors format. You can [download them](/davanstrien/nasa_concept_art/tree/main) in the Files & versions tab.