Bytedance Seedream V4.5 Edit Sequential

Seedream V4.5 Edit Sequential is the "give me a batch of variations from one edit prompt" node in the Seedream family. You give it a single prompt, optionally some reference images, and it produces 1 to 15 edited images with cross-image consistency - the whole point of the "Sequential" line is that the batch hangs together rather than being 15 unrelated takes. It's the Seedream V4.5 edit model with its typography chops, but aimed at volume: mood boards, option sheets, poster-variant sets.

The "sequential" part does real work: the images are generated in sequence with the previous results informing the next, which is what buys you the coherence. It's priced per image (~$0.027/image on the sequential line), so a 15-image run is a real purchase - plan the count before you queue, not after. This is a "drafts for review" node, not a "generate one perfect poster" node; for a single keeper, the plain Edit node is the cheaper route.

How it works

The node takes your prompt and max_images (1–15, default 4), sends them to WaveSpeed's Seedream V4.5 Edit Sequential endpoint, and returns a batched IMAGE output - all generated images in one tensor, ready to split with an Image Batch / Split batch node or preview as a grid. Two automation details worth knowing:

• max_images is appended to the prompt automatically - the tooltip says the node injects the image count into your prompt for API compliance, so you don't have to.
• image_url is optional - unlike the single-edit nodes, you can run this with no source images for prompt-to-multiple generation, or pass up to 10 reference URLs.

The inputs that matter

• prompt - the edit description. Since it's a batch, describe the set: "variations of this product shot with different background colors" reads better than a single-target instruction.
• max_images - how many coherent results you want, 1–15.
• image_url - optional reference image(s), up to 10.
• size_preset / width / height - resolution (1024–4096px here); custom dims only with the "Custom" preset.
• seed - cache control; randomize re-runs the whole batch (and re-bills it).

enable_sync_mode and enable_base64_output are the API-scripting extras, and show_aspect_ratio is cosmetic.

Installing it

Part of the ERPK Collection (eRepublik-Labs/comfyui-nodes-erpk). ComfyUI Manager: search erpk → ERPK Custom Nodes → restart. Or:

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

WaveSpeed API key required (Settings → ERPK → API Keys). No model downloads.

Common issues

• The batch got expensive fast - max_images × per-image price. For iteration, start at 2–4; scale up only when the prompt is dialed in.
• Images come out as one tensor - that's normal for the output. Split the batch or preview the grid; don't expect 15 separate image sockets.
• Reference URLs must be public - WaveSpeed fetches them server-side. Localhost or expiring links fail.
• Text in a batch - V4.5 handles typography well, but small text at low resolution still garbles; keep the outputs big enough to read.