Update March 23, 2026: Luma AI has released its new image generation model, Uni-1, which has garnered significant attention in human preference tests conducted by Luma Labs. According to the results, Uni-1 ranks first in overall performance, style/editing, and reference-based generation categories. In contrast to its competitor, Google’s Nano Banana, Uni-1 secured the second position in pure text-to-image generation capabilities.
The performance of Uni-1 on a benchmark prompt suggests it performs comparably to Google’s flagship Nano Banana Pro, potentially even surpassing it. This marks a notable improvement over Midjourney v8, which had difficulty handling the same prompt. It is important to note that the images evaluated were processed through a Luma image generation agent; hence, results may vary slightly when the model is accessed via the upcoming API. Users can explore Uni-1 at no cost through Luma Labs.
In terms of pricing, Uni-1 competes favorably with its rivals while delivering comparable resolution. The upcoming API is expected to offer images at a resolution of 2K for an average cost of approximately $0.09 each, which varies depending on the number of reference images used.
| Feature | Uni-1 | Nano Banana 2 | Nano Banana Pro |
|---|---|---|---|
| Text to Image (2048px) | $0.0909 | $0.101 | $0.134 |
| Image edit / i2i (2048px) | $0.0933 | $0.101 | $0.134 |
| Multi-ref, 1 img (2048px) | $0.0933 | $0.101 | $0.134 |
| Multi-ref, 2 imgs (2048px) | $0.0957 | $0.101 | $0.134 |
| Multi-ref, 8 imgs (2048px) | $0.1101 | $0.101 | $0.134 |
While Nano Banana 2 offers lower resolutions at reduced prices, with a 0.5K image costing about $0.045 and a 1K image priced around $0.067, Uni-1’s features appear to justify its pricing model.
Original article from March 8, 2026: Luma AI has unveiled Uni-1, a groundbreaking model that integrates both image understanding and generation within a single framework. Like its counterparts, Google’s Nano Banana Pro and GPT Image 1.5, Uni-1 employs an autoregressive transformer. This advanced model generates content sequentially, token by token, rather than relying on traditional diffusion techniques that generate images from noise. Both text and images are processed through a unified pipeline, enhancing the overall efficiency of content generation.
Luma AI asserts that Uni-1 is capable of reasoning through prompts dynamically during the generation process. This functionality allows the model to deconstruct complex instructions effectively, resulting in more precise adherence to user prompts. One of the model’s standout features is its ability to amalgamate several images into a cohesive new composition, demonstrating its advanced logic capabilities.
In one notable demonstration, Uni-1 produced an entire sequence from a single reference image, depicting a pianist aging from childhood to old age while maintaining consistent scene composition and camera angle. Such capabilities are expected to enhance creative workflows in various applications.
Uni-1 excels in logical image processing, achieving the highest score on the RISEBench test among competitors, including Nano Banana 2 and GPT Image 1.5. Its enhanced image generation capability significantly bolsters its visual understanding, allowing it to nearly match the object recognition capabilities of Google’s Gemini 3 Pro. Luma AI indicates that Uni-1 is capable of processing multiple languages, broadening its accessibility.
With plans to make Uni-1 available through Luma Agents and the Luma API, the market is poised for increased competition in AI-generated imagery. While specific pricing details have yet to be released, the introduction of Uni-1 signals a new era for AI-assisted creative tools, potentially reshaping industries reliant on visual content generation.
See also
Sam Altman Praises ChatGPT for Improved Em Dash Handling
AI Country Song Fails to Top Billboard Chart Amid Viral Buzz
GPT-5.1 and Claude 4.5 Sonnet Personality Showdown: A Comprehensive Test
Rethink Your Presentations with OnlyOffice: A Free PowerPoint Alternative
OpenAI Enhances ChatGPT with Em-Dash Personalization Feature
