ByteDance Seed

ByteDance Seed is the company's flagship LLM line, and this node runs Seed 2.0 through BytePlus's ModelArk API - with a twist that makes it stand out in this family: it takes not just images but actual videos as context. If you've ever wanted to ask a model "what changed between these two clips" or "describe this footage for a caption," this is the node for it. It's the same ByteDance that feeds you Seedream for images and Seedance for video; the company runs a two-track posture where the flagship products stay API-only, and this node is the front door to the LLM track.

The mechanism

Send a prompt, optionally attach up to 20 images and 4 videos, and the node uploads that media to Comfy's storage, builds a Responses API request, and returns a STRING. Like every node in this family it routes through Comfy's proxy and bills through your Comfy account in credits - no ByteDance/BytePlus key of your own, price badge on the node. Landed in ComfyUI core in May 2026, so it's one of the newer partner nodes.

The inputs that matter

• prompt - the text, multiline.
• model - a dynamic combo of Seed 2.0 Pro, Lite, and Mini. Pick a model and the node grows the matching inputs: images (up to 20), videos (up to 4), and an advanced temperature slider (0.0 to 2.0). Lite and Mini are the budget picks; Pro is the flagship.
• seed - default 0. Read the tooltip before you trust it: the seed just controls whether the node re-runs. Results are non-deterministic regardless. It's a re-run trigger, not a reproducibility promise.
• system_prompt - optional foundational instructions.

Output is a single STRING - same story as the rest of the family: text display, text encode, save, or feed it onward.

Getting it and what to watch

Ships with ComfyUI core, nothing to install. The catches are the family ones: it needs a Comfy account with credits, it needs internet, and every run costs tokens. Seed's pricing badge is notably cheap - this is one of the budget-friendly providers in the partner lineup - but the media is what multiplies the bill, and video especially.

Where people trip up: the 4-video cap is hard - more clips and the node throws an error, so pick your best shots. And don't chase a "seed" for reproducibility here; if you need deterministic output, this isn't the node. Finally, it's cloud-only, so if your workflow must stay fully offline, reach for a local LLM pack instead. But if you've got clips and questions, this is the shortest path from footage to a useful answer.