NewBie-image-Exp0.1 - An open-source image generation model from NewBieAi Lab

What is NewBie-image-Exp0.1?

NewBie-image-Exp0.1 is an open-source experimental image generation model from NewBieAi Lab, based on the Next-DiT architecture with 3.5B parameters. The model was pre-trained on the full dan + 1m e621 dataset and trained for four months on 8×h200 hardware, accumulating approximately 23,000 hours of training. It uses Google/Gemma3-4b-it and Jina Ai/Jina Clip v2 as text encoders, paired with Flux 1 Dev-VAE. NewBie-image-Exp0.1 focuses on high-quality anime image generation, producing anime images with rich detail and striking visual effects.

Main functions of NewBie-image-Exp0.1

• High-quality anime image generationBased on training with a large amount of high-quality anime data, it can generate anime-style images with rich details and striking visual effects.
• Strong architectural foundationIt adopts the Next-DiT architecture, with 3.5B parameters, and has powerful image generation capabilities.
• High-efficiency training and optimization: Use high-performance computing hardware for long-term training to ensure model performance and stability.
• Flexible text encoderCombines Google/Gemma3-4b-it and Jina AI/Jina Clip v2 to improve the matching accuracy of text understanding and image generation.
• Potential for multi-scenario applicationsIt is applicable to fields such as animation creation and game design, providing powerful tool support for creative workers.

Technical Principles of NewBie-image-Exp0.1

• Based on Next-DiT architectureThe model employs the advanced Next-DiT architecture, a deep learning architecture designed for efficient generation of high-quality images and capable of handling complex image generation tasks.
• Large-scale data pre-trainingThe model was pre-trained using the full dan + 1m e621 dataset, which contains a large number of high-quality anime images, enabling the model to learn rich image features and styles.
• High-performance computing supportThe model was trained for four months on high-performance hardware with 8×h200 processors, accumulating approximately 23,000 h200 hours, ensuring that it was fully trained on large-scale data to improve the generation results.
• Text encoder fusionBy combining Google/Gemma3-4b-it and Jina AI/Jina Clip v2 as text encoders, the model's ability to understand text descriptions is enhanced, and images that meet the text requirements are generated more accurately.
• Optimized VAE configurationThe Flux 1 Dev-VAE is used for image encoding and decoding to optimize the quality and detail of the generated images and improve the overall generation effect.
• Deep learning and generative adversarial networksBy utilizing deep learning techniques, especially the principles of Generative Adversarial Networks (GANs), the quality of generated images is continuously optimized through adversarial training between the generator and the discriminator, making them more realistic and diverse.

The project address for NewBie-image-Exp0.1

• Github repository: https://github.com/NewBieAI-Lab/NewBie-image-Exp0.1
• Huggingface model libraryLink: https://huggingface.co/NewBie-AI/NewBie-image-Exp0.1

Application scenarios of NewBie-image-Exp0.1

• Animation CreationIt provides powerful image generation tools for animation artists, enabling them to quickly generate high-quality animation characters and scenes, thus accelerating the creative process.
• Game DesignUsed for character design, scene construction, and concept art creation in game development, helping designers quickly realize their ideas.
• Digital ArtIt supports digital artists in creating unique anime-style artworks, providing a wealth of visual materials and sources of inspiration.
• Illustration and comicsIt assists illustrators and comic artists in generating illustrations and comic sketches, improving creative efficiency and quality.
• Advertising and MarketingIn the advertising and marketing field, it is used to generate attractive anime-style promotional images to enhance visual appeal.
• Education and TrainingAs a teaching tool, it helps students and beginners learn anime drawing techniques, providing reference images and creative inspiration.