Seedream5.0 Pro - Text to Image

The flagship tier, hosted

Seedream 5.0 Pro's generator, exposed as a node. Prompt in, image out, ByteDance's best closed image model on the other end of the wire. There is no local Seedream - the family stays API-only while ByteDance ships the components underneath as open weights - so if this model's output is what you want, a node like this is the only path to it.

Worth knowing what you're buying, because it's counterintuitive: this node's quality field gives you 1K at basic and 2K at high. The Lite sibling advertises 2K/3K/4K. Pro charges more for a stronger model at lower resolution ceilings. That's a deliberate product split - Pro for compositional quality and edit fidelity, Lite for output size on a budget - and if you came here expecting a 4K toggle you'll be confused until you read the tooltip.

Inputs

• prompt - required, 3 to 5000 characters. Stock text is a café-design-tool blurb; replace it. The generous limit is an invitation to describe layout and text placement, which is where this family is strongest - graphic and typographic composition rather than skin texture.
• aspect_ratio - required, default 1:1.
• quality - required: basic → 1K, high → 2K.
• output_format - png or jpeg.
• nsfw_checker - default False on this node (it defaults True on the Pro image-to-image sibling, so verify which one you're looking at).
• timeout_seconds - 1200 by default, up to 7200, in 30-second steps. callback_url - leave empty.

Outputs: image, url, all_urls_json, task_id, raw_json, credits_consumed, credits_left.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/felipederosilva/ComfyUI-KIE-Nodes-Next
pip install -r ComfyUI-KIE-Nodes-Next/requirements.txt

Restart ComfyUI, then paste the KIE key once at Settings → KIE.ai Nodes Next → Connection. The pack verifies it against KIE before overwriting an existing key, stores it outside your workflows, and shows your balance alongside the connected status. Only requests, Pillow, numpy and PyYAML get installed.

Practical notes

Text is the known weak spot even on Pro: for a single simple element it's fine, and for dense layouts testers still report blur and drift - roughly the level of the better hosted generators, not a typography engine. If your composition depends on clean type, generate the layout and add type locally, which is also cheaper.

Cost discipline matters more here than on any other node in this batch, simply because Pro is the premium tier. Use the Lite node for exploration and this one for the shot you're actually delivering, one change per reroll. The node reports credits spent and remaining balance on itself after every generation, so a runaway loop is visible rather than theoretical.

And the standard API-node caveat: your prompt and any reference you attach leave your machine, and the provider's moderation runs regardless of how you set nsfw_checker. It's a filter you can't patch, because there are no weights to patch.