Seedream 4.0: Nano Banana Killer?

AI image wars continue, with bonus rumors of an upcoming OpenAI image update.

TL;DR

ByteDance’s Seedream 4.0 gives Nano Banana a run for its money with top-tier image generation and editing.

What is it?

Just when you thought we were done, folks!

It’s only been two weeks since Google’s Nano Banana (Gemini 2.5 Flash Image) “killed” Photoshop, but we already have a shiny new kid on the block: Seedream 4.0.

The latest model from ByteDance goes toe-to-toe with Nano Banana on image editing tasks, while also being on par with GPT image generation at making new images.

Source: ByteDance

This video has plenty of helpful side-by-side image editing tests:

As you can see, Seedream 4.0 handles many niche challenges better than either GPT image generation or Nano Banana (often both).

How do you use it?

Right now, there’s no official free way to try Seedream 4.0.¹

Update 15-09-2025: You can now use Seedream 4.0 for free on LMArena:

Simply follow this guide I wrote for Nano Banana but select seedream-4 or seedream-4-high-res from the dropdown:

Several shady third-party sites claim to offer free Seedream 4.0 generations, but the ones I tested all use outdated models under the hood. ByteDance’s China-facing platform jimeng.ai does offer free daily Seedream 4.0 credits, but I couldn’t make it work without a local telephone number.

I expect ByteDance might soon add Seedream 4.0 to its Dreamina site.²

For now, if you’re outside China, your best option is a paid subscription to one of the following:

• krea.ia

• freepik.com

If you don’t want to commit to a subscription, fal.ai lets you try Seedream 4.0 on a pay-per-use basis (3 cents per image):

1. Navigate to fal.ai

2. Sign up (or sign in with e.g. your Google or GitHub account)

3. Purchase credits over at https://fal.ai/dashboard/usage-billing/credits

4. Pick the Seedream 4.0 model you want to use:

   1. For image generation:
      https://fal.ai/models/fal-ai/bytedance/seedream/v4/text-to-image

   2. For image editing:
      https://fal.ai/models/fal-ai/bytedance/seedream/v4/edit

5. Type in your prompt and click “Run” to generate or edit the image

   

I used the long test prompt from my GPT-4o showdown article:

Candid photo of a purple steampunk platypus with wings on stage in a comedy club. To the left of the platypus is a cyberpunk goose playing a saxophone. Behind them is a show banner with the words “Top Billed.” In front of the stage, in the audience, is a dieselpunk duck. The duck is holding a handwritten show program that says “Welcome to Top Billed! No beaks were harmed in the making of this lineup. Prepare for laughs, feathers, and unpaid sax solos.”

Here’s the best of four attempts by Seedream 4.0:

It sticks to the prompt and almost nails the text (apart from the last word). Here’s a side-by-side view of Seedream 4.0, Nano Banana, and GPT-4o—click any image to open it in full resolution:

Left to right: Seedream 4.0, Nano Banana, GPT-4o Image

Seedream 4.0 is also the only model trying to honor the “candid photo” look.

Take Seedream 4.0 for a spin with your own prompts, especially those you felt were too tricky for other image models. See if it’s an improvement.

Why should you care?

ByteDance has just joined the ranks of OpenAI and Google with a context-aware image model that has built-in reasoning abilities. This means Seedream 4.0 doesn’t just blindly follow instructions but truly understands the request and has the world knowledge to act on it.

So I can ask it for something as broadly defined as this:

Four-panel cartoon that explains how coffee makes it to the shop

And it might return something like this:

With Seedream 4.0, China now has a top-tier contender in yet another GenAI field, which reminds me of when Kling and Hailuo AI came seemingly out of nowhere to dominate AI video at the end of 2024. (Several months later, reasoning models had their DeepSeek R1 moment.)

Every time I feel we’re near the peak of what’s possible with AI image generation, a new model comes out to raise the bar.

And it looks like we’re not quite done for this year, because OpenAI is allegedly gearing up to update its own image model, fueling wild speculation:

Source: X

As someone who originally started Why Try AI because of AI image models, I’m excited to see where it’s all heading.

🫵 Over to you…

Have you tried Seedream 4.0 yet? Are you following the AI image field closely? What is your favorite image model or platform?

Leave a comment or drop me a line at whytryai@substack.com.

---

Thanks for reading!

If you enjoy my writing, here’s how you can help:

• ❤️Like this post if it resonates with you.

• 🔄Share it to help others discover this newsletter.

• 🗣️Comment below—I love hearing your opinions.

Why Try AI is a passion project, and I’m grateful to those who help keep it going. If you’d like to support my work and unlock cool perks, consider a paid subscription:

1. Let me know if you manage to make it work for you.

2. Dreamina lets you use Seedream 3.1, so it’s likely that the new version will also be added.