seedream

Seedream is ByteDance's closed image model, and the honest truth is you cannot download it, ever - no weights, no local checkpoint, no abliteration workaround. It's an API-only model. So this node isn't competing with your local Flux or SDXL setup; it's the way you bolt a frontier cloud model that would otherwise live on a website into your actual ComfyUI graph. The moment you accept that trade (per-call cost, your prompt leaves the machine), it's a genuinely useful node.

What it is

One of ~160 nodes in Civitai's official civitai-comfy-nodes pack. Like every recipe node in the pack, it doesn't run anything locally - it packages your prompt and settings, submits a job to Civitai's Orchestration API, polls until it's done, and drops the finished image back as a native IMAGE tensor. From the canvas it looks like a normal generator. Underneath it's an HTTP client with your Civitai credentials baked in, and every job is billed in Buzz on your Civitai account.

The node's own version dropdown is the fun part: v3, v4, v4.5, or v5.0-lite. ByteDance has been shipping Seedream revisions fast, and the pack generates its node shapes straight from the orchestration API spec - so when a new Seedream lands, the version list updates without you reinstalling anything.

Inputs and outputs that matter

The only required input is prompt. The rest are optional and most have sane defaults:

• version - pick your Seedream revision. Defaults to empty (the API's default), but if you want the newest, set it explicitly.
• guidance_scale - defaults to 2.5, range 1–10. Lower is more faithful to the prompt; higher is more literal/contrasty. Rarely worth much fiddling.
• width / height - 256–4096, default 1024². Seedream handles big canvases natively, which is one of its selling points.
• quantity - 1–12 images per run.
• images - yes, there's an IMAGE input hiding in the optional list. This node quietly does image-to-image too: feed a reference image and Seedream works from it. That's a bigger deal than the marketing admits, because Seedream's edit/reframe abilities are part of why people pay for it.
• enable_safety_checker - off by default; flip it on if you want the checker (and you should understand the checker is Civitai/ByteDance's policy, not yours).
• seed, output_format, image_metadata - standard.

Outputs are the pack's uniform set: images (IMAGE), errors (STRING), workflow_id and raw_json (both for debugging and, usefully, checking what a job cost).

Install

cd ComfyUI/custom_nodes
git clone https://github.com/civitai/civitai-comfy-nodes.git
pip install -r civitai-comfy-nodes/requirements.txt

Or ComfyUI Manager → search "Civitai Comfy Nodes" → Install → restart. You'll also want a Civitai Model Selector node wired into nothing here - Seedream has no model socket, so you don't need one. Auth is the one thing you can't skip: set CIVITAI_API_TOKEN (headless), or connect via the sidebar / a Civitai Auth node.

Gotchas

This is a closed model running on someone's servers, so all the standard cloud caveats apply: it's slow compared to local (a job is seconds-to-minutes plus queue), it costs Buzz per image, and it's filtered at the source - there is no community bypass for a model with no weights. Civitai's version of Seedream also runs its own safety machinery, which is why enable_safety_checker and the moderation outputs exist elsewhere in the pack.

If a job fails, check errors and raw_json on the node - a 400 there is usually an unsupported resolution for the version you picked. And remember: this is an early-preview pack, so version lists and behavior can shift between releases. For a model you can't run locally, that's a fair trade.