Bytedance Seedream V4 Sequential

Seedream V4 is ByteDance's earlier generation of its image model - the one before V4.5's typography overhaul - and the Sequential variant is its batch superpower: one prompt, 1 to 15 images, kept consistent with each other. It's the budget storyboard tool: character sheets, mood boards, multiple angles of the same subject, all in a single billed call at $0.027 per image.

It runs through the WaveSpeed AI API via the ERPK Collection, the community pack wrapping the service for ComfyUI. No local model; the cloud does the consistency math.

How it works

Write one prompt, set max_images, and the node handles the rest - including silently appending the image count to your prompt for the API ("…, 4 images"), which is worth knowing so you don't fight it or double-write it. The result comes back as an images IMAGE tensor: a batch, not a single frame. Feed it to a batch-aware save/preview node to get all N images out.

The resolution range here is wider than V4.5's - 320 to 4096px, default 1408×1408 - so it scales from small quick batches up to large single-style output.

The inputs that matter

• prompt - the concept for the whole batch. Consistency is anchored in the prompt, so describe the character or subject the same way the whole family should look.
• max_images - 1–15, default 4. Auto-appended to the prompt; don't add the number yourself.
• size_preset - presets or Custom, with width/height 320–4096 (default 1408) when Custom.
• seed - cache control, randomizes by default. Fixed seed reuses a cached batch; randomize for a fresh, re-billed generation.
• enable_sync_mode / enable_base64_output - sync-wait and base64 options, ignorable at first.

Install

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

Or ComfyUI Manager → search erpk → ERPK Custom Nodes. WaveSpeed key with credits via right-click canvas > ERPK Settings (Settings > ERPK > API Keys).

Common issues

• Batch looks consistent-ish but not identical - that's the realistic bar for sequential generation. Faces and props drift across large batches; keep the prompt tight and the batch small.
• "Why is there a number in my prompt?" - the auto-append. Intentional, leave it.
• Saving a batch - the images output is multi-frame; a single-image Save node will complain. Use a batch-aware save node or split the batch first.
• Cost - $0.027/image × 15 is real money. Batch at 4–6 while exploring, then scale when the direction is locked.

The honest take: if your output needs legible text, step up to the V4.5 Sequential node in this pack. If it's a character sheet or a style family with no typography involved, this is the cheaper and perfectly good choice. The sequential siblings are otherwise twins: same max_images, same $0.027/image, same auto-append quirk.