Bytedance Seedream V5.0 Lite Sequential

What it is

This node runs ByteDance's Seedream V5.0 Lite - a closed, API-only text-to-image model - from inside ComfyUI, via WaveSpeed's API. The "Sequential" part is the point: instead of one image, it generates a whole batch (1–15) with cross-image consistency. Same character, same style, same scene logic across the set. That's your character sheets, product lineups, book-cover series, or "show me the same subject in 12 outfits" requests - the thing you'd otherwise hack together with a fixed seed and hope.

There's a real reason to use the Lite tier specifically: it's Seedream's budget option, trading down from the full V5.0 for a lower per-image price, and it enforces a higher minimum resolution (1440px) than the V4.x nodes.

How it works

Under the hood it's a single API call, not a local diffusion run. The node takes your prompt, appends "Generate a set of N consecutive." to it, sends max_images and your resolution to WaveSpeed, downloads the returned image URLs, and hands you a ComfyUI IMAGE batch. Your GPU isn't involved at all - no model to download, no VRAM to budget. ByteDance keeps Seedream API-only while it gives the open-source world its components; this node is just a clean front door to the paid product.

The inputs that actually matter

Most of the schema you can ignore. These four do the work:

• prompt - required. Leave it empty and the node throws a ValueError before anything is billed. This is the entire creative control; be specific about the subject and the consistency you want across the set.
• max_images - 1 to 15, default 4. Every image is billed separately.
• size_preset - recommended resolutions like 1:1 (2048x2048), 16:9 (2560x1440), up to 1:1 4K. Leave on Custom and use width/height instead (1440–4096, default 2048×2048, step 8) if you want something in between.
• seed - this one is a trap, see below.

The output is a single images output (an IMAGE batch), wired straight into SaveImage or PreviewImage.

The seed trick that saves you money

WaveSpeed's endpoint has no real API seed. So the seed widget is repurposed for cache control: leave it on randomize (-1) and every queue re-runs the API call - and re-bills you. Fix a value and ComfyUI sees the same fingerprint, serves the cached result, and skips the call entirely.

Consequence: if you're iterating on a prompt you like, fix the seed so your tweaks to other parts of the workflow don't burn a fresh generation each time. If you want genuinely new images every run, keep it randomized and accept the charge.

Installing the pack

1. ComfyUI Manager → Install Custom Nodes → search erpk → ERPK Custom Nodes → install → restart.
2. Or:

cd /path/to/ComfyUI/custom_nodes/
git clone https://github.com/eRepublik-Labs/comfyui-nodes-erpk.git erpk
cd erpk
pip install -r requirements.txt

Then set your WaveSpeed key once in right-click canvas → ERPK Settings → API Keys. That's it - no weights, no checkpoints.

Cost and gotchas

• ~$0.035 per image per the pack's docs, so a default 4-image batch runs about 14¢ and a maxed 15-image batch about 52¢. Watch max_images; it's the most expensive slider on the node.
• Pricing and model behavior drift, because WaveSpeed resells Seedream - check their current per-image rate before committing to a big batch.
• enable_base64_output (default off) makes the node return base64 instead of URLs, for consumers that can't fetch URLs. Leave it off unless you know you need it.
• enable_sync_mode (default off) waits for generation before returning; off means it returns when the request is accepted. For interactive work you usually want it on so the images actually show up.