Seedream 4

This is the flagship node of the FloyoAI Seed API pack, and it earns the display name "Seedream 4" - it's the one you'll actually reach for. Seedream 4 is ByteDance's current image generation model, an API product in the same way all of Seedream has always been (see the ByteDance KB panel: flagship products closed, components open). The node is a remote control for that API, with a twist that's rare in ComfyUI: its image inputs appear and disappear as you connect things.

No images connected means text-to-image. Connect one image and it's image-to-image editing. Connect two or more and it becomes multi-image blending. The node's JavaScript watches your connections and auto-adds image slots up to a hard cap of ten, renumbering them as you go. It's genuinely slick - you don't pick a mode, the graph's shape picks it.

How it works

The mechanism underneath is one model (seedream-4-0-250828) and one endpoint, but three batch modes with very different behavior:

• single - one API call, one image. Uses your seed, or random at -1.
• parallel - fires batch_size concurrent API calls via a thread pool, each with a random seed. You get N variations of the same prompt in roughly the time of one. Best for exploring looks.
• sequential - one API call with sequential_image_generation: auto, producing thematically related images that share a narrative. Good for storyboard-ish sets. With input images, total input + output must stay at or under 15.

Your images are base64-encoded and sent inline - no external hosting, no upload step, which the README rightly brags about. The response images are downloaded and stacked into a single batched IMAGE tensor.

Inputs and outputs that matter

• prompt - the driver in every mode; describe style, lighting, composition.
• width / height - output size, defaults 2048×2048, up to 4096.
• batch_mode and batch_size - see above. batch_size caps at 15.
• image1 (plus auto-added image slots) and seed.

Output is one IMAGE tensor holding however many images the mode produced - feed it to Save Image, or use batch-splitting nodes if you want each variation separately.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/FloyoAI/ComfyUI-Seed-API
cd ComfyUI-Seed-API
pip install -r requirements.txt

Restart, then put your BytePlus API key in config.ini or set BYTEPLUS_API_KEY. You can also find it in ComfyUI Manager under "ComfyUI Seed API Integration".

Where people get burned

Two traps. First, a blank black image is the pack's error fallback - the console shows the real error, and with a cloud API that's usually "model not enabled for this key" or a quota/credit problem. Second, parallel mode ignores your seed on purpose; every request gets a random one for variety. If you need reproducible results, use single or sequential. And don't treat sequential as "more of the same" - the API intentionally varies the images, which is a feature unless you expected a batch of near-identical renders. Price-wise, each image is a paid API call, so a 15-image sequential run adds up; check the console's usage log if you're surprised by the bill.