SF WaveSpeed Seedream V4

Seedream is ByteDance's flagship image line, and it has never been open-weighted - the entire Seedream family is API-only, full stop. So if you want Seedream V4's distinctive high-res output in ComfyUI, you're going through a hosted API, and SF WaveSpeed Seedream V4 is the node that does it: prompt in, up-to-4096-px image out, no weights to download, no GPU needed.

The thing that makes Seedream worth trying is its aggressive native resolution. The size_preset list goes straight to 2048×2048 as the default and caps at a real 4096×4096 - not upscaled-to-look-big, but generated at that size. For large-format print work or images you plan to crop hard, that's a genuine advantage over 1K-class hosted models.

It's part of the comfyui-stillfront pack, under Stillfront/WaveSpeed.

How it works

The node posts your prompt and size preset to WaveSpeed's bytedance/seedream-v4 endpoint, waits for generation, and converts the returned image URL into an IMAGE tensor. Simple on your side - the model lives entirely on WaveSpeed's servers.

The inputs that matter:

• prompt - the image description.
• size_preset - nine options, all high-res: 2048x2048 (1:1) default, then 4:3/3:4, 16:9/9:16, 3:2/2:3, an ultra-wide 3024x1296 (21:9), and 4096x4096 (1:1) at the top end. Resolution is the point of this model, so this is the dial that defines your job.
• seed - default 0; set a fixed seed for reproducible output.
• enable_sync_mode - and here's a trap: the default is off, unlike most of the other nodes in this pack. That means the API returns a task id and the node has to poll. It does still wait for the result and hand back the image, but expect the "submitted successfully" console message and a longer pause before output. If you want the tighter sync path, flip it on.

Output is a single output_image (IMAGE).

Installing it

Pack install plus WaveSpeed key:

cd ComfyUI/custom_nodes
git clone https://github.com/razvanmatei-sf/comfyui-stillfront
cd comfyui-stillfront
pip install -r requirements.txt

Restart, then wire in an SF WaveSpeed Client node with your key (or config.ini / WAVESPEED_API_KEY).

Troubleshooting

• Generation takes a while / nothing happens for a minute. Async mode is on by default. Check the ComfyUI console for the task id line - it's working, just polling.
• 4096×4096 is slow and expensive. It's generating four times the pixels of a 2K image, so expect four times the time and roughly four times the bill. Do drafts at 2048, spend on the final.
• "API key is empty." Client node → config.ini → WAVESPEED_API_KEY, in that order.
• Community note on the filter. WaveSpeed's web UI is noticeably stricter than its API - the API is the path "which supports broader content use cases," but it's still a hosted service with its own terms, so keep your commercial obligations in mind.

Where it sits: if Nano Banana is Google's hosted flagship and Qwen is Alibaba's, Seedream is the ByteDance entry - and thanks to its native high resolution, it's the one this pack's image nodes where big output is the default rather than the exception.