In China’s domestic AI image generation market, most products follow the same path: they use Stable Diffusion as the underlying engine and build interfaces, curated models, and prompt optimization on top of it. While this approach lowers technical barriers and allows for rapid product launches, it also means relying entirely on overseas open-source communities from the ground up, lacking true technological independence.
Tiamat chose a different path. Developed by an AI startup based in Shanghai, this tool claims to have trained its own image generation model from scratch—not a fine-tuned version of Stable Diffusion, but a system with an independent technical architecture. The product is named after Tiamat, the Babylonian goddess of chaos; the metaphor of creating order from nothingness and chaos fits AI art generation quite well.
What Is Tiamat?
Tiamat (tiamat.world) is one of the few domestic AI image tools that truly pursues an independent R&D path. Many competing domestic products are essentially wrappers around Stable Diffusion, unable to bypass their dependence on overseas open-source communities. Tiamat chose to train its own image generation model from scratch, ensuring full autonomy in technical iteration without relying on SD version updates or community models.
The product initially operated in a closed beta phase, gradually opening up to creator communities. Feedback from early beta users indicates that its generation quality, particularly regarding character details, ranks among the higher tiers for domestic self-developed tools.
The Significance of the Technical Path
Why does independent development matter? On the surface, for ordinary users, there is little perceptible difference between using SD and a self-developed model; generating an image is just generating an image. However, this choice has profound implications across several dimensions:
Autonomy over Copyright and Licensing: Stable Diffusion uses the CreativeML Open RAIL-M license, which imposes certain constraints on commercial use, while controversies surrounding training data copyright persist. A self-developed model can establish clearer legal standing regarding data sources and training methods, which is crucial for commercial products.
Complete Control Over Model Direction: Within the SD ecosystem, a model’s foundational capabilities are constrained by upstream version iterations; sometimes, desired features require waiting for community implementation. With an independent path, the team determines training directions autonomously, allowing for deep optimization tailored to specific styles or scenarios.
Independence from Overseas Communities: The SD community is primarily overseas. Given China’s specific regulatory requirements for AI-generated content, products relying entirely on overseas models face compliance uncertainties. Self-developed models allow for better content safety measures and localization adaptations.
Key Features
Generation Quality
Tiamat demonstrates competence in both realistic styles and artistic illustration. Character generation is the capability most users care about; details such as facial proportions, skin texture, and hair strands often serve as the litmus test for a model’s quality. Based on early beta feedback, Tiamat’s performance in this area stands out among domestic self-developed tools.
It also covers landscapes, architecture, and abstract art styles, though generation quality may vary across different styles. Actual testing is recommended to judge specific results.
Chinese Prompt Support
This is a common advantage for all domestic AI image tools and a foundational capability of Tiamat. Compared to foreign tools that require constructing complex prompts in English, describing visual ideas directly in Chinese is more intuitive and allows for more natural expression of nuanced emotions and atmosphere.
Creative Community
Tiamat features a community centered around AI art creation, where users can publish works and exchange creative insights. This is a plus for users who value feedback and inspiration during the creative process—a good AI art community is not just about showcasing final products but also sharing the creative process and prompt strategies.
Continuous Iteration
An independent R&D path means the team has the capacity to drive technical iterations at its own pace. As training data accumulates, algorithms are optimized, and computing power investment increases, the speed and direction of improvements for a self-developed model can be controlled autonomously. This offers a more stable development trajectory compared to products dependent on upstream models.
Comparison with Major Competitors
Compared to Midjourney: MJ remains the industry benchmark for AI image quality, difficult to surpass in aesthetic style and overall visual effect. However, MJ requires Discord operation, access to overseas networks, lacks a Chinese interface, and often suffers from unstable access within China. Tiamat provides a smoother user experience within the domestic environment, making it a practical choice for those unwilling to deal with overseas tools.
Compared to Wenxin Yige (Baidu): Both are domestically self-developed AI image tools. Wenxin Yige benefits from Baidu’s technical accumulation in its ERNIE Bot and significant resource investment, boasting higher brand recognition and a larger user base. As a startup product, Tiamat has its own characteristics in certain style directions; the two are not mutually exclusive, and trying both is advisable.
Compared to SD-based Domestic Tools (e.g., 360 Zhihui, Yijian AI): The greatest advantage of the SD ecosystem is its richness of models—hundreds of fine-tuned models from the community cover almost every style with dedicated options. Tiamat’s independent path lacks the early-stage ecological richness of SD-based tools but holds greater initiative in technological independence and future development direction.
Compared to Local Stable Diffusion Deployment: The core advantages of local SD are free, unlimited generation, offline usability, and an extremely rich model ecosystem, making it a strong choice for technically skilled users. Tiamat’s advantage lies in its low barrier to entry—it requires no local environment configuration or knowledge of SD parameters and workflows; it is ready to use out of the box, suitable for ordinary users who do not wish to tinker with technology.
Who Is It For?
Creators interested in the development of domestic AI image technology: If you care about the technical path behind the tool, Tiamat’s independent direction is a choice for understanding and supporting the independent development of domestic AI technology.
Users needing stable access within China: If you want to avoid dealing with overseas network access issues and unstable overseas services, and need an AI image tool that works reliably in the domestic environment, Tiamat is a worthy option.
Individual users interested in AI art creation: Tiamat’s community atmosphere and Chinese support are quite friendly to Chinese-speaking users who enjoy exchanging ideas and sharing work.
Users wanting to compare different tools: If you are already using Midjourney or SD-related tools and want to understand the effects and differences of domestic self-developed paths, Tiamat is a valuable reference point for comparison.
Limitations
The independent R&D path comes with costs. Compared to Stable Diffusion, which has years of development and a massive community ecosystem behind it, Tiamat lags in the diversity of model styles and the richness of community resources. The SD ecosystem has countless fine-tuned models tailored to different styles; such accumulation cannot be replicated in the short term.
As a product from a startup, the sustainability of its commercialization is also a factor users must consider—whether a tool is easy to use is one thing, but whether the product can be maintained long-term is another. The stability of domestic AI startup products varies widely, which is background information worth understanding before use.
Access and Pricing
Tiamat initially operated in a closed beta phase, accessible via application or invitation codes. As the product matures, its accessibility has gradually expanded. For the current access status and registration methods, please refer to the latest information on the official website.
It offers free quotas, with paid plans unlocking more generation counts and advanced features. Pricing strategies are subject to the official website; startup products typically adjust their pricing as they progress through different stages.
Tiamat represents a technical direction in China’s AI image field worth observing. In an industry context where everyone relies on overseas open-source models, pursuing independent R&D is a difficult but valuable attempt. Whether it can catch up to or even surpass overseas tools in generation quality and ecosystem building remains to be seen over time.