Proxima Aurelia 2

More Than a Prompt Follower: The Aesthetic Intelligence of Proxima Aurelia 2

Proxima Aurelia 2 doesn't just follow your prompts — it understands them visually. Built on Krea 2, a foundation image model developed completely from scratch with a singular focus on aesthetics, style transfer, and creative control, this model represents a fundamental shift in how AI approaches image generation. Where conventional models converge toward a safe, polished "AI look," Proxima Aurelia 2 was trained to span an extraordinary range of visual languages, from the grainiest film photography to the cleanest digital illustration. Its architecture — a diffusion transformer (DiT) enhanced with grouped-query attention, sigmoid-gated attention, and Qwen3-VL text encoding — was built from the ground up to treat style as something you guide, mix, and adjust, not just a vague keyword in a prompt.

Available exclusively on Proxima.art, Proxima Aurelia 2 brings a level of creative sophistication that transforms even the simplest text descriptions into visually compelling compositions. The model's training pipeline — spanning pretraining, midtraining, supervised fine-tuning, preference optimization, and reinforcement learning — ensures that every generation benefits from a deep, nuanced understanding of visual aesthetics. It currently ranks among the top 10 models on the Artificial Analysis text-to-image leaderboard and places second among independent labs, proving that aesthetic ambition and technical excellence can coexist.

Five Visual Strengths That Define Proxima Aurelia 2

• Exceptional Aesthetic Sensibility: Proxima Aurelia 2 was trained with an explicit focus on aesthetics. Its multi-stage training pipeline — which includes dedicated midtraining on curated stylistic domains and supervised fine-tuning on hand-picked visual datasets — means the model develops an innate sense of what makes an image visually compelling, not just technically accurate.
• Intelligent Composition: Beyond rendering objects correctly, Proxima Aurelia 2 understands visual hierarchy, balance, and framing. The model's prompt expander, trained through a two-stage SFT and RL pipeline, maps simple or underspecified prompts into richer visual directions, automatically suggesting compositional choices that enhance the final image without overwriting your intent.
• Rich Visual Storytelling: From the simplest prompt, Proxima Aurelia 2 generates images that feel like they belong to a larger narrative. Its training on densely captioned images — where captions describe not just what's present but how elements relate to each other — imbues the model with an understanding of mood, context, and visual storytelling that goes far beyond surface-level rendering.
• Authentic Style Transfer: The model's style-reference system reads the style components of reference images and transfers them into your output with minimal content leakage. You can combine multiple style references, adjust how strongly each influences the result, and control variation across a batch — giving you a level of creative control that feels like working in a design tool rather than a chat interface.
• Text Rendering and World Knowledge: Proxima Aurelia 2 incorporates OCR-extracted text into its training captions and leverages Qwen3-VL for enhanced visual understanding, enabling it to render text within images and reference known entities accurately — a capability that proves invaluable for signage, labels, and concept illustrations that demand factual precision.

Where Proxima Aurelia 2 Excels: Four Scenarios Worth Exploring

Proxima Aurelia 2 shines brightest when the brief calls for more than just accurate object placement. Here are four scenarios where its unique capabilities deliver standout results:

• Concept Art and Fantasy Illustration: Whether you're visualizing a character, a creature, or an entire fantasy landscape, Proxima Aurelia 2's understanding of mood, lighting, and compositional drama brings your ideas to life with a painterly richness that feels both imaginative and cohesive. The model handles complex digital painting styles and experimental directions with equal confidence.
• Social Media Content with a Distinctive Voice: From Instagram carousels to blog header images, Proxima Aurelia 2 can maintain a consistent aesthetic across a series of generations — or deliberately vary them for visual interest. Its built-in LoRA styles let you shift between pink Barbie glamour, surreal dreamscapes, and 80s horror nostalgia with a single click, making it ideal for creators who need visual variety without starting from scratch each time.
• Marketing and Advertising Visuals: The model's luxury-oriented aesthetic sense and ability to render opulent, radiant compositions make it particularly well-suited for premium product photography, fashion lookbooks, and lifestyle imagery. Its text rendering capabilities ensure that brand names, product details, and promotional copy appear crisp and legible within the generated image.
• Personal Artistic Exploration: If you're looking to experiment with styles you can't easily produce yourself — grainy film photography, vintage cinematography, surreal abstract visuals — Proxima Aurelia 2's style-reference system lets you upload a reference image and have the model internalize its visual language. Combine multiple references, adjust their influence weights, and discover unexpected visual directions you might not have imagined.

Getting the Most Out of Proxima Aurelia 2: Settings and Prompting

Proxima Aurelia 2 is built on the Krea 2 architecture, a diffusion transformer (DiT) model that responds particularly well to structured prompting and deliberate parameter tuning. Here are our recommended starting points:

• CFG Scale / Guidance Scale: Start around 7–10. Proxima Aurelia 2's aesthetic intelligence means it doesn't typically require the high guidance values that other models need to follow prompts — pushing it past 12 can introduce artifacts and flatten the very stylistic richness that makes it distinctive.
• Sampling Steps: 30–50 steps with a high-quality sampler (DPM++ 2M Karras or Euler A) strikes the best balance between generation speed and visual fidelity. The model's architecture includes lightweight timestep modulation that helps maintain quality even at moderate step counts.
• Resolution and Aspect Ratio: Proxima Aurelia 2 supports generation at multiple resolutions, with particularly strong results at 1024×1024 (square) and 1024×576 (16:9 landscape). For portrait-oriented content, 576×1024 works well. The model was trained with a curriculum that progressively scaled resolution from 256px to 1024px, so it handles higher resolutions with greater stability than models trained only at a single resolution.
• Prompting Strategy: Proxima Aurelia 2 includes a built-in prompt expander that automatically enriches simple prompts with visual detail. This means you can start with something as brief as "a warrior at sunset" and let the model add compositional and atmospheric detail. However, if you have a specific vision, be explicit about mood, lighting direction, and camera angle — the model responds exceptionally well to these cues. Avoid overloading prompts with contradictory style descriptors; the model handles 2–3 style references better than 5+ competing directions.

Tip: The single most impactful setting for Proxima Aurelia 2 is the style-reference weight. When using reference images, start with a moderate influence (around 40–50%) and adjust based on whether you want the reference to guide the overall aesthetic or merely inspire a direction. Lower weights preserve more of your original prompt intent; higher weights push the output closer to the reference's visual language.

Six Built-in LoRA Styles: Instant Aesthetic Shifts on Proxima.art

One of Proxima Aurelia 2's most practical features is its suite of six built-in LoRA styles, available directly on Proxima.art. These LoRAs activate automatically when selected, letting you shift the model's visual personality without retraining or complex configuration. Each LoRA has been calibrated to work harmoniously with the base model — no conflicting style weights or unexpected artifacts.

• Pink Barbie (weight 15): This LoRA infuses your images with a cohesive pink-toned aesthetic, wrapping everything — from the subject and clothing to the background and lighting — in a palette of soft pinks, rose golds, and blush accents. It's perfect for Barbie-inspired fashion photography, feminine product visuals, or any project where you want pink to dominate without feeling monotonous. Lean into it fully for a stylized editorial look, or keep it subtle for a softer, more natural pink undertone.
• Proxima Dream (weight 15): A surreal, ethereal, and otherworldly atmosphere awaits with this LoRA. It blends elements of fantasy, imagination, and abstract visuals to capture the feeling of being in a dream where reality bends and boundaries blur. Ideal for concept art, album covers, or any project that calls for a departure from literal representation. The style works especially well with soft lighting, pastel color palettes, and subjects that suggest movement or transformation.
• Luxury Sparkle (weight 10): Embodies elegance, opulence, and radiant beauty, combining high-end aesthetics with dazzling, light-reflecting elements. This style is perfect for generating images that exude sophistication, glamour, and a sense of exclusivity — think jewelry photography, luxury fashion campaigns, or glamorous portraiture. The LoRA adds subtle sparkle highlights and rich tonal depth that elevate the perceived value of any subject.
• Horror 1980x (weight 20): Captures the eerie, nostalgic, and suspenseful essence of classic 80s horror films. This style blends the aesthetic of grainy VHS tapes and vintage cinematography with the dark, unsettling atmosphere of slasher films and psychological thrillers. The LoRA introduces film grain, chromatic aberration, and a desaturated yet high-contrast color grade that instantly evokes the look of VHS-era horror. Best used for horror-themed content, retro-futuristic projects, or any image that benefits from a gritty, nostalgic edge.
• Grim Darker Glamour 80s (weight 7): A darker, more brooding take on 80s glamour. While the standard 80s glamour LoRA leans into neon brightness and celebratory energy, this variant introduces shadow, grit, and a more melancholic undertone. It's ideal for noir-inspired photography, cyberpunk aesthetics with a darker edge, or character portraits that demand emotional depth and visual tension. The lower default weight means it blends subtly — consider increasing it for a more pronounced effect.
• Detail Control (weight 320): Unlike the other LoRAs, this one isn't a style but a flexible detail slider. It lets you increase or reduce image complexity while preserving the original style of your generation. Push it higher for intricate textures, fine details, and dense visual information — perfect for architectural visualizations or detailed character renders. Pull it back for cleaner, more minimalist compositions where the overall shape and mood matter more than pixel-level detail. The wide weight range (up to 320) gives you granular control over how busy or serene your final image feels.

Available Now on Proxima.art

Proxima Aurelia 2 is available for image generation on Proxima.art, where you can access all six built-in LoRA styles, experiment with style references, and explore the model's full creative range. Whether you're generating a single image or running a batch of exploratory renders, the platform provides a streamlined interface that puts Proxima Aurelia 2's aesthetic intelligence at your fingertips. Visit the site to see current pricing and start creating.

The Next Step in Aesthetic Image Generation

Proxima Aurelia 2 represents a meaningful departure from the generic "AI look" that has dominated text-to-image generation. By building aesthetic intelligence into the model itself — through careful data curation, a multi-stage training pipeline, and a style-reference system that treats style as something you guide rather than merely request — it offers a generation experience that feels genuinely creative rather than merely responsive. Whether you're a professional designer looking for a reliable aesthetic engine, a content creator needing distinctive visuals at scale, or an artist exploring new visual directions, Proxima Aurelia 2 on Proxima.art gives you the tools to move beyond the expected and into something that feels unmistakably yours.