ByteDance claims that Seedream 4.0 beat Gemini 2.5 Flash Image for image generation and editing on its internal evaluation benchmark. — SCMP
TikTok-owner ByteDance has launched its latest image generation artificial intelligence tool Seedream 4.0, which it said surpasses Google DeepMind’s viral “Nano Banana” AI image editor across several key indicators.
ByteDance’s Seed department, which oversees the tech unicorn’s AI development efforts, said on Tuesday that Seedream 4.0 had strong image editing capabilities, putting it in direct competition with Nano Banana, officially known as Gemini 2.5 Flash Image.
The US model has received wide acclaim since its release in late August for the consistency and accuracy of its image editing abilities, which have been a challenge for AI image generators until now.
ByteDance claims that Seedream 4.0 beat Gemini 2.5 Flash Image for image generation and editing on its internal evaluation benchmark MagicBench, with stronger performance in prompt adherence, alignment and aesthetics.
However, these results were not published as part of an official technical report.
According to Artificial Analysis, a leading AI benchmark firm, Seedream 4.0 represented a “significant evolution” from previous ByteDance tools because it combined the text-to-image capabilities of Seedream 3.0 with the image editing capabilities of SeedEdit 3.0, while still being priced the same as the former at US$30 per 1,000 generations.
Gemini 2.5 Flash Image currently tops the firm’s leader boards for both text-to-image generation and image editing while Seedream 4.0 has yet to be evaluated.
Seedream 3.0 is currently ranked fifth for text-to-image generation and sixth for image editing.
Compared to earlier models, Seedream 4.0 has a new architecture that speeds up raw image inference speed by over 10 times, making it a much faster tool to use, the Beijing-headquartered tech giant said.
Feedback has been positive online, as users praised the accuracy of the editing function, which can make quick changes to images through text prompts.
The tool has been made available to domestic users on ByteDance’s Jimeng and Doubao AI apps, as well as to corporate clients through its official cloud service, Volcano Engine.
On Fal.ai, a global generative media hosting platform, Seedream 4.0 costs US$0.03 per generated image, while Gemini 2.5 Flash Image is priced at US$0.039.
China has been quick to embrace AI-powered image and video generators, with the field hotly contested among major tech firms such as short-video app operator Kuaishou Technology and video gaming giant Tencent Holdings.
The state has signalled its support for AI-generated content by recognising their copyright in late 2023, but has also recently introduced mandatory labelling of such content.
The field has seen a wave of new products in recent years offering more competitive prices, including video generation tool Vidu, jointly developed by Tsinghua University and Beijing-based start-up Shengshu AI.
On Tuesday, Vidu’s developers rolled out a new reference-to-image function internationally that generates images based on a combination of up to seven reference images, priced at 0.62 yuan (US$0.09) per image. Gemini 2.5 Flash Image allows up to nine references. – South China Morning Post
