The Atomeaser is a style trained on nine surrealist painters from various schools: René Magritte, Salvador Dalí, Max Ernst, Giorgio de Chirico, H.R. Giger, Maurits Cornelis Escher, Zdzisław Beksiński, Dariusz Zawadzki and Wayne Douglas Barlowe. It aims to provide a global style incorporating all the ideas of those painters into a single brush. It will excel at drawing surreal scenes from both the deepest nightmares to the highest dreams.

1. Understand the Dataset and Model

• The LoRA model has been trained on 225 surreal captions, with descriptions that combine abstract and surrealist themes.

• The captions tend to focus more on visual elements, mood, and environment, often with minimal explicit object-person interactions.

• The captions describe environments with figures, objects, and surreal structures, without emphasizing strong narrative interactions.

2. Key Insights from the Captions

• Focus on Atmosphere and Objects: The captions frequently mention static objects and environmental features rather than actions or events. This suggests your prompts should emphasize surreal objects, structures, and landscapes.

  • Examples of frequent nouns: "figure," "scene," "structure," "foreground," "humanoid."

• Abstract Descriptions: There are many references to abstract features like color palettes (e.g., "blue," "red," "browns") and tones. Leverage these by prompting the model with vague or abstract ideas, like "dreamlike landscapes with humanoid figures and otherworldly structures."

• Minimal Object-Person Interactions: The model is not heavily trained on interactions between figures and objects. You can prompt for static relationships like “figures standing beside mechanical structures” rather than expecting dynamic actions (e.g., "holding," "reaching").

• Average Sentiment Polarity: The dataset is nearly neutral in sentiment, so prompts don't need to push extremes like "horrifying" or "beautiful." Focus on ambiguity and unease, key to surrealism and subtle tension.

3. Prompting Tips for Best Results

• Use Surreal and Abstract Terms: Incorporate terms related to surrealism like "ethereal," "dreamlike," "mechanical," "organic," or "distorted" to tap into the model’s training on these themes.

• Focus on Visual Elements: Mention color, shapes, and positions of objects and figures. Use phrases like:

  • “A humanoid figure surrounded by floating geometric shapes.”

  • “Mechanical structures intertwined with organic forms in a barren landscape.”

  • “A figure stands before an expansive horizon with distorted perspectives.”

• Leverage Ambiguity: Don’t over-specify. Surrealism thrives on ambiguity and vagueness, so give the model space to create its own interpretation of abstract concepts. Examples:

  • “A mysterious figure gazing into an endless void of muted colors.”

  • “Strange objects suspended in mid-air with no apparent logic or gravity.”

• Incorporate Known Themes: Use common elements from the 9 painters that influenced the model’s training. These can include:

  • Mechanical and futuristic elements (Giger, Beksinski)

  • Otherworldly figures and landscapes (Barlowe, Dali)

  • Distorted perspectives and geometries (Escher, Chirico)

4. Prompt Example Variations

• Static Focus: “A large humanoid figure standing in the foreground, with intricate structures in the background and a vibrant blue sky above.”

• Abstract Focus: “Geometric shapes float aimlessly in an infinite expanse, creating a sense of surreal disorientation.”

• Color and Palette: “A dreamlike scene filled with tones of red and blue, where figures blend into the background of intricate machinery.”