mahnoor·fatima

Slim Aarons Generative Studio

Three generated photographs in a mid-century holiday idiom: a domed poolside pavilion, figures on the terrace of a pale villa, and a swimming pool shaded by a leaning tree, each captioned with its prompt.

A style-preservation and reinterpretation workflow using LoRA fine-tuning.

This project develops a reproducible pipeline for preserving and extending Slim Aarons’s mid-century photographic aesthetic through ethically transparent generative AI workflows.

Overview

A curated dataset of high-resolution Slim Aarons photographs was used to train a LoRA that captures compositional structure, color palette, lighting, and subject framing without replicating specific originals.

The trained model was deployed through ComfyUI workflows and a Gradio web app, enabling accessible experimentation across text-to-image, image-to-image, editing, animation, and 2D-to-3D outputs.

Methodology and System Design

Data Curation

  • 150 high-resolution images selected across eras and locations
  • Each image paired with a detailed descriptive caption
  • Captions encoded composition, color relationships, lighting, subject positioning, and mood

Prompt and Trigger

A standardized caption structure was introduced, along with a dedicated trigger token:

SLMRNS. A Slim Aarons photograph of…

This ensured consistent invocation of the learned style during generation, and provided a clear marker of stylistic conditioning.

Three generated images shown as result cards, a rooftop terrace with clipped hedges, a patio of pink cushioned seating, and a laid outdoor table, each listed with its seed, step count and full prompt.
Every result kept with the seed and prompt that produced it, so any output can be reproduced.

Model Strategy

  • Base model: FLUX.1-dev
  • Fine-tuning method: LoRA
  • Rationale: modular, lightweight style injection without altering the base model

Training Setup

  • Resolution: 1024 by 1024
  • Hardware: a single A100 GPU
  • Roughly 2,000 training steps
  • Tuned LoRA rank and text encoder parameters
  • Periodic checkpoints and sample review for quality assurance

Quality Control

Outputs were evaluated during training for:

  • Composition accuracy
  • Color fidelity
  • Architectural and clothing detail
  • Consistency across varied prompts

The final LoRA was exported as a compact safetensors file for reliable reuse.

An input and output comparison. On the left a photograph of a glass residential tower, in the middle the settings used, FLUX.1-dev with the Slim Aarons LoRA, trigger SLMRNS, transformation strength 1.0, guidance scale 2.5 and 28 inference steps, and on the right two versions of the same tower restyled as luxury apartments in the learned idiom.
The same building put through the model at full LoRA strength.

Gradio Web App

A browser-based interface provides:

  • Multiple generation modes
  • Sliders for LoRA strength, guidance scale and denoise strength
  • Seed control for reproducibility

This enables use by non-technical users in teaching and research contexts.

The web app's cinemagraph tab, headed create a cinemagraph or story from an image and a descriptive prompt, with a drag and drop upload area and three collapsed prompt sections beneath.
Turning a still into a short loop.
The web app's filter and animation panel, a dropdown of effect types open over a list of options, above a button to apply the chosen style filter and create an animation.
Style filters and animation, applied to the last generated image.

Additional tools generate:

  • GIFs and cinemagraphs
  • Ken Burns style pan and zoom
  • Multi-frame storyboard sequences

These extend still imagery into short-form narrative outputs.

Ethics and Provenance

  • Outputs explicitly marked as AI-generated
  • Dataset provenance documented
  • Internal trigger token retained for traceability
  • Project framed as educational and preservational, not substitutive

The project delivers a reproducible Slim Aarons Generative Studio supporting research, teaching, and creative exploration.

Planned extensions include usability testing, rights assessment for broader release, Hugging Face demo deployment, and expansion toward responsibly curated style-preservation toolkits.