Metanomaly K.K., the Tokyo-based operator of the global AI 2D creation platform PixAI, has officially released Tsubaki.3, its next-generation anime image generation model. The launch follows about one month of early access and shifts attention from creating a single polished image to developing existing characters and ideas into new settings, scenes, and manga. The model introduces smart reference inputs, natural language editing, a color palette feature, and improved multi-character scene composition for creators worldwide.
Higher Native Resolution and Stronger Core Performance
Tsubaki.3 supports native image generation up to 2048 by 2048 pixels, approximately four megapixels. This is four times the total pixel count of the previous Tsubaki.2 model at 1024 by 1024 resolution. PixAI says the upgrade enables denser detail in characters, costumes, materials, and backgrounds while improving prompt fidelity, complex composition, lighting control, style range, and fine elements such as hair and lace.
Smart Reference for Character Continuity
The most significant change in Tsubaki.3 is its focus on what happens after a character is first generated. Creators can use up to three reference images through the smart reference system without building or training a separate dataset. From a single character image, users can produce expression sheets, costume variations, poses, turnaround views, and age changes while keeping the character's major visual traits relatively stable across results.
Editing Without Restarting the Image
Tsubaki.3 strengthens natural language image editing so users can refine an existing result instead of repeatedly generating a full image. After uploading an image, creators can describe which areas to modify, what to replace, and which elements to preserve. The feature supports changes such as hair color or outfit adjustments while maintaining pose, background, and lighting, and it can also fix hand gestures, add or remove objects, and edit text inside an image.
Manga Pages and Complex Scene Composition
The model extends beyond single images by composing manga pages with multiple panels, speech bubbles, dialogue, screen tones, speed lines, close-ups, and wide shots. Users can describe a story moment in a text prompt and use a previously created character image to build scenes featuring that character. PixAI emphasizes support for story development tasks such as character settings, storyboards, panel layouts, and manga pages rather than fully automated long-form webtoon production.
Video Models and Launch Promotion
Alongside Tsubaki.3, PixAI released Tsubaki Video and Tsubaki Video Flash to extend characters and scenes into moving content. Both video models are available to all users regardless of membership, and generation costs are discounted by 30 percent until October 20. The company says the Tsubaki product family is designed to carry one character and idea across images, manga, and video rather than treating each format as an isolated experience.
Global Contest and Platform Momentum
To mark the official launch, PixAI is holding a global creative contest until October 20 with no restrictions on style or theme. Participants can submit works made with Tsubaki.3 through the PixAI platform and official X account, with prizes that include PixAI credits and Amazon gift cards. PixAI has registered more than 16 million users worldwide since its October 2022 launch and continues to grow in markets such as South Korea, North America, and Taiwan.
Tsubaki.3 represents PixAI's strategic expansion from generating one strong image to supporting a continuous reference-edit-develop workflow for character-centered creation. By connecting image generation, editing, manga composition, and video in a unified model family, the platform aims to help creators preserve and evolve their ideas across formats. PixAI states that the key improvement is the creative process after the first image, supported by higher resolution, stronger scene control, and more flexible editing tools.