Gemini’s 2.5 Flash Image model, also known as Nano Banana, has seen its popularity skyrocket thanks to the recent 3D model trend that has taken hold of the internet. While Gemini already has a lot of competition from the likes of ChatGPT, Qwen, Grok, and others, the chatbot is now challenged by another image-generating model from TikTok’s parent company, Bytedance.
The new model, Seedream 4.0, has overtaken Gemini Nano Banana on Artificial Analysis—a leaderboard that ranks different LLMs on metrics like quality, speed, and cost. Seedream 4.0 holds the first spot, while Nano Banana is now in second place, with OpenAI’s GPT-4o in fourth and Alibaba’s Qwen Image Edit in fifth.
What is Seedream 4.0? Why is it special?
Seedream 4.0 is capable of not only generating images from natural language text prompts but also making precise edits to existing images, similar to what Gemini Nano Banana has been doing.
Seedream 4.0 claims to extract key details from reference images, including character design, artistic style, and product features, which enables tasks such as character creation, style transfer, and product design. The model is also capable of accepting multiple image inputs to help users combine these images.
Seedream 4.0 vs. Gemini Nano Banana:
We tested Gemini Nano Banana and Seedream 4.0 with five different prompts to check which model is able to generate better images. All of the images in this comparison were generated using LMArena’s side-by-side comparison feature.
Prompt 1:
Basic 3D model
“Create a 1/7 scale commercialized figurine of the characters in the picture, in a realistic style, in a real environment. The figurine is placed on a computer desk. The figurine has a round transparent acrylic base, with no text on the base. The content on the computer screen is a 3D modeling process of this figurine. Next to the computer screen is a toy packaging box, designed in a style reminiscent of high-quality collectible figures, printed with original artwork. The packaging features two-dimensional flat illustrations.”
In this image, Seedream managed to retain the facial expressions from the original image, but it did not create an actual figurine like Gemini did.
Prompt 2:
City-Explorer 3D model
“Use the supplied photo as the primary reference for face, hair and posture. Generate a 1/7-scale, fixed-pose collectible figurine (high-quality PVC/resin look) of the subject in modern streetwear—backpack, coffee cup in hand—standing on a round transparent acrylic base (no text or logos). Place the figurine on a real wooden computer desk. To the left, a widescreen monitor shows the 3D modelling process of this very figurine (wireframe viewport, material/shader node editor, object hierarchy and a small thumbnail of the reference photo). To the right, a premium toy packaging box sits upright—high-quality collectible style with flat 2D illustrations (orthographic front/side/back line art), matt cardboard texture and spot varnish on the photo panel. Lighting: warm desk lamp key, soft overhead fill, subtle rim light; camera: 35mm full-frame equivalent, f/2.0, shallow depth of field so the figurine is tack-sharp and background soft. Render photoreal, true skin tones, micro-detail on fabric, realistic paint finish on sculpted details. No text on the acrylic base; no watermarks.”
Gemini managed to retain enough of the facial details and posture from the original image to make it look like a 3D model of me. Meanwhile, Seedream 4.0’s images were of higher quality, but the representative model bore little resemblance to the image uploaded by me.
Prompt 3:
Urban Samurai crossover:
“Using the provided image for face and stance, produce a 1/7-scale collectible that blends modern streetwear with subtle samurai armour elements (textile jacket panels with lacquered arm guards, sheathed katana). Place on a transparent round acrylic base (no text) set on a home workspace. The computer screen displays the modelling process (skeleton / rig view + sculpt passes) and a packaging box features elegant flat 2D character illustrations in three views printed with a soft-touch cardboard look. Lighting: moody city evening via window + soft desk lamp; camera 35mm, f/2.0, emphasise fabric micro-details and lacquer sheen. Photoreal, editorial composition suited to social thumbnails.”
Prompt 4:
Renaissance painting
“What would this look like as a renaissance painting?”
While Gemini did a pretty decent job of turning a selfie of mine into a renaissance painting, Seedream did not complete the task despite multiple attempts.
Gemini wasn’t perfect either; while it managed to keep the facial features of the celebrity intact, it completely messed up my facial features but retained the clothes I wore in the selfie.
Prompt 5:
Turn objects into hologram
“Turn the charger into a 3D transparent line art hologram”
Both Gemini and Seedream did a decent job on this image, but Nano Banana once again showed its supremacy by retaining more details and overall keeping the image more realistic.
Overall, I feel Gemini Nano Banana still reigns as the incumbent champion in the image editing department. However, it does tend to make some mistakes, like losing facial details or lowering the resolution of the ultimate output.
