SEEDREAM 4.0: Tiktok creators head on Google Banana for the image generation

It is not new: the world of generative artificial intelligence is perpetual boiling. And more particularly when one of his main representatives makes an important announcement, the others were quick to answer him. Also, after the shock wave caused at the end of August byGoogle Bananait’s the turn of another tech giant to get into the ring. Bytedance has just unveiled Seedream 4.0, its new model of generation and image editing. An announcement which obviously sounds like a direct response to the Google tool, whose qualities in terms of image editing quickly went around social networks. The ambition is clear: carving out a share of this market in full effervescence.

Unified architecture

On paper, the main innovation of Seedream 4.0 lies in its architecture. Rather than offering two separate solutions, Bytedance has chosen to merge its previous technologies into a single system. SEEDREAM 4.0 thus combines image generation capacities from text by Seedream 3.0 with SeedEdit 3.0’s editing features. But this merger is not just a grouping. According to the brand, the goal here is to streamline the creative process for users, who no longer have to juggle between different tools to create an image and then refine it.

In parallel, Bytedance also announces a speed of execution more than ten times greater than that of the previous version. The tool would thus be able to generate 2K resolution images in less than two seconds. This speed is welcome to allow rapid iterations, without the usual long waiting times.

Technical capacities revised upwards

Beyond the speed, the features have also been expanded. SEEDREAM 4.0 can now produce images up to a resolution of 4K, with flexible management of formats. What can we really do with exactly? To start, the tool allows you to make retouching with completely natural language controls. A simple sentence is therefore enough to request to delete an object from a photo, adjust the lighting or transform an existing image by applying a different style.

To guarantee better visual consistency, the model can also rely on six reference images in order to maintain a constant style on all of the creations. The rendering of the finest elements such as text, tables or diagrams has also been improved according to Bytedance. The brand highlights fairly varied use cases, ranging from the design of graphics to illustrate climatic data to the creation of historical chronological friezes or vintage -style websites.

The (supposed) match match

Inevitably, who says competitor says comparison. But we reject all artificial intelligence that must be taken with huge tweezers the data communicated when launching new models. Without too much surprise, Bytedance says that his model beat Gemini 2.5 Flash Image (the official name of Nano Banana) during tests carried out internally. According to the said house test bench, called Magicbench, Seedream 4.0 would be superior in terms of respect for the instructions (the famous prompt adhesion), consistency and aesthetic quality.

But these results from an internal evaluation, prudence is even more stake. Especially since for the time being, on public and independent rankings like that of the artificial analysis firm, it is the Google model that retains first place, both for the generation of images and for publishing. The Battle of benchmarks is just beginning. We also notice with a hint of fun that more grandmundle only talks about the Chatgpt image generatorwhile he was a great out of a few months ago.

Aggressive pricing

To win and be widely used, we know that the only performance is not always enough for a model. Accessibility and price are also determining factors. Bytedance deploys its tool on several fronts. In China, it is integrated directly into consumer applications such as Doubao and Jimeng. For companies and developers, access is via the Cloud Volcano Engine platform. Globally, Seedream 4.0 is also available via third -party partners such as Fal.ai, Freepik or Wavespeed.ai.

In terms of prices, the strategy seems aggressive. On the Fal.ai platform, an image generation with Seedream 4.0 is billed 0.03 dollars, against 0.039 dollars for its Google rival. A minimal gap, but which could weigh in the scale for large volume users. On Volcano Engine, the price even drops to around $ 0.028 per image.

With such pricing, Seedream 4.0 can undoubtedly shade the Google solution, provided that the qualitative gap is not too important.

Numériques settles in Beaugrenelle Paris for The most tech days : product demonstrations, use or purchase advice, exchanges with our journalists … Discover the full program.