aquif-Image-14B

A high-performance text-to-image generation model with 14.3B parameters, fine-tuned from Wan 2.2 A14B for enhanced image quality across photorealism, spatial reasoning, creativity, and portrait generation.

Demos

Graffiti tag reading "aquif-Image-14B" sprayed across a weathered concrete wall. Hyper-detailed photorealism. Crisp paint edges with subtle overspray. Rich texture in the wall, chipped layers, grime, faint moisture streaks. Natural daylight. Slight shadows from surrounding urban elements. Depth, scale, and tactile realism, as if captured with a high-resolution street-photography lens.	A candid portrait of a woman leaning against a bus window, soft afternoon light, slight half-smile, faint eye crinkle, believable skin texture, tiny flyaways in her hair catching the light, street scene blurred behind the glass.	A scratched stainless steel thermos resting on a wooden workbench, diffuse studio lighting, micro-scratches visible, faint fingerprints on the metal, grain variation in the wood, shallow depth of field.
A skateboarder mid kickflip above a sunlit ramp, accurate body posture, board rotation captured mid-air, dust motes frozen, crisp motion arcs implied through pose rather than blur, shadows matching direction of light.	A cluttered reading nook with overlapping objects, a stack of books partially covering a ceramic mug, a lamp casting warm spill light across uneven surfaces, fabric folds on a chair, believable object layering.	A research whiteboard filled with formulas, diagrams, arrows, variable names, and small typed labels, clear handwriting, consistent perspective, ink density variation, natural smudges around erased areas.

Evaluation

Metric	aquif-Image-14B	Flux.2 [dev] 32B	HunyuanImage-3.0 80B	Qwen-Image 20B	Ovis-Image 7B	Wan 2.2 A14B
Photorrealism	1104	1114	1092	1083	1083	1076
Spatial Reasoning	1053	1050	1047	1024	1038	1043
Creativity	1099	1132	1143	1134	1092	1079
Portraits	1121	1140	1123	1074	1080	1067
Text Readability	1116	1132	1053	1101	1120	1060

Substantial improvements over baseline Wan 2.2 Image across all metrics: +28 photorealism, +10 spatial reasoning, +20 creativity, +54 portraits, +56 text readability.

Model Details

• Parameters: 14.3B
• Base Model: Wan 2.2 A14B
• Architecture: Diffusion-based text-to-image
• License: Apache 2.0
• Training: Fine-tuned for enhanced image generation quality

Installation

pip install diffusers transformers torch

Use Cases

• Creative content creation and concept art
• Product visualization and marketing assets
• Portrait and character design
• Architectural and interior visualization
• Game asset generation
• Research and technical illustration
• Accessible professional image generation

Roadmap

• aquif-Image-5B: Lightweight 5B parameter variant based on Wan 2.2 5B (coming soon)

Acknowledgements

• Wan Team: Base model (Wan 2.2 A14B)
• HuggingFace: Diffusers library
• aquif AI Research Team: Fine-tuning and optimization

---

Made in 🇧🇷

© 2025 aquif AI. All rights reserved.