Bytedance Seedream V5.0 Lite

Seedream V5.0 Lite is a text-to-image node that runs on ByteDance's Seedream model over the WaveSpeed API - which means it's ByteDance's flagship image quality, and the only thing your machine has to do is hold a URL. It's the cheapest route into the Seedream V5 generation: the Lite tier starts at a 1440px minimum resolution and prices around $0.035 per image, which is notably cheaper than the standard Seedream line.

Here's the context that makes this node make sense: ByteDance keeps its flagship generation models closed - Seedream and Seedance are API products, full stop (the open releases from the same lab are the research-y infrastructure like PuLID and Depth Anything). So there is no "just download Seedream V5 and run it locally." If you want that ByteDance look, an API node like this is the only game in town. The tradeoff is that you're paying per image, so it's a tool for when you need the result, not for burning through a thousand variations on a sampler that's free.

How it works

The node sends your prompt to WaveSpeed's Seedream V5.0 Lite endpoint, waits for the render, and pulls the returned image URL down into a real ComfyUI IMAGE tensor. Because it returns a tensor, it drops straight into your existing image chain - a VAE decode step isn't needed (there's no latent, this is not a diffusion sampler), but any downstream image node will accept it happily.

The resolution presets are the interesting part. V5.0 Lite's minimum is 1440px, which is a step up from the older Seedream tiers - the model is built to hold a lot of detail, and it's tuned for typography, so posters, UI mockups, and anything with rendered text is where it shines. You get a size_preset dropdown with 1:1, 4:3, 3:2, 16:9, and their portrait/4K variants, or "Custom" to set width/height yourself (1440–4096px).

The inputs that matter

• prompt - the description. Seedream is good at reading long, specific prompts, so be generous.
• size_preset - the presets (like 1:1 2048×2048 or 16:9 2560×1440) or Custom for manual control.
• width / height - only used when size_preset is "Custom", so don't bother with them otherwise.
• seed - cache control only; randomize re-runs generation (and re-bills), a fixed value lets ComfyUI reuse the cached result.

The optional client input wires to a WaveSpeed API Client node; leave it unconnected and the key comes from Settings. show_aspect_ratio just displays the ratio in the node title - harmless, cosmetic.

Installing it

This ships in the ERPK Collection (eRepublik-Labs/comfyui-nodes-erpk). Install via ComfyUI Manager (search erpk → ERPK Custom Nodes) or manually:

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

You need a WaveSpeed API key (Settings → ERPK → API Keys, or wavespeed/config.ini). No model downloads - the pack's WaveSpeed dependency is just pydantic, requests, and Pillow for tensor conversion.

Common issues

• "Prompt is required" - the node refuses blank prompts. Fill it in.
• No image URLs in result - generation failed or the account ran out of credit. Check your WaveSpeed balance; it bills per render.
• Watch the resolution math - widths/heights snap to 8px steps and have a 1440px floor on this model. If you type something below the floor it gets rejected by the API.
• It's billed - randomize seed = new charge per run. For iteration, that adds up fast.