Bytedance Seedream V4

Seedream V4 is the flagship image generation node in the pack's WaveSpeed section: ByteDance's Seedream V4, served as a hosted API. Type a prompt, get an IMAGE tensor back - and the range of output sizes is the headline. From 320×320 up to 4096×4096, this is one of the few hosted image models you can push to true 4K without an upscale pass, which puts it in a different league than the pack's other hosted generators on raw resolution.

The ByteDance context matters for expectations: Seedream is an API-only product line, full stop. The KB's ByteDance panel is explicit that the entire Seedream and Seedance lines are hosted products with no open weights. So this node is a paid wrapper, and the value proposition is "ByteDance's best image model without writing a line of API code," not "free model, small wrapper."

The inputs

• prompt - the image description
• size_preset - the recommended resolution presets; select Custom to use the manual width/height fields
• seed - for cache control, randomizes by default
• width / height - 320 to 4096 (step 8), only used when size_preset is Custom
• show_aspect_ratio - displays the aspect ratio in the node title
• client - optional WaveSpeed client

The output is a single image IMAGE tensor - downloaded for you, graph-ready, no URL juggling. Wire it to Preview, Save Image, or an upscaler.

The design details worth knowing

size_preset vs manual width/height is the main structure to grasp. The presets are the recommended sizes (fastest path to good results), and the Custom option unlocks the raw 320–4096 range. If you're printing or compositing at high res, that Custom slider going to 4096 is the feature - but mind the bill: bigger = more pixels = more money per call, and the README's per-image pricing for the Seedream line sits around $0.027/image, with sequential multi-image variants costed the same per image. Iterate at preset sizes, render final at Custom.

The seed widget is cache control, like the other WaveSpeed image nodes - the README is explicit that these endpoints have no API seed, so the value isn't sent; a fixed value lets ComfyUI serve a cached result and skip re-billing. Leave it randomizing when you want a fresh, billed generation.

Install and setup

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

Or "ERPK Custom Nodes" from ComfyUI Manager, restart, and set a WaveSpeed API key in ERPK Settings (right-click canvas). No model downloads - it's all hosted.

Troubleshooting

Two realistic failure modes. Empty prompt: the node validates before billing, so you'll get a prompt-required error rather than a wasted call. Size out of range: width/height below 320 or above 4096 (or not a multiple of 8) gets rejected server-side - stick to step 8. And one expectation-setting note: Seedream V4 is strong on photorealism and scene coherence, but it's a commercial service with content moderation, so it is not the tool for the uncensored end of the hobby. If that's the use case, you're better served by local open weights - the KB's ByteDance panel notes the open/closed split exists for exactly this reason.