Bytedance Seedream V5.0 Lite Edit Sequential

What it is

The edit-flavored sibling of the Seedream V5.0 Lite Sequential node. Same idea - one remote API call, 1–15 output images with cross-image consistency - except instead of generating from text, it edits images you already have. Feed it up to 10 source images plus a text description of the change, and you get back a coherent batch where the same modification (same character, same style shift, same lighting treatment) has been applied across every input.

That's the workflow for "take my 8 product shots and re-render them all as watercolor," or "these 6 character sketches, now in the same cinematic style." Doing that with local models means a per-image edit pass plus a consistency fight; here it's one call.

How it works

Mechanically it's the Sequential node with one extra input: image_url. Your prompt describes the modifications, the node appends "Generate a set of N consecutive." to satisfy the API, and the source images ride along as URLs. Because it's Seedream V5.0 Lite through WaveSpeed, it enforces the same higher 1440px minimum resolution as its text-to-image sibling - the source images are resized into that band rather than whatever resolution they arrived at.

The results come back as a single images IMAGE batch for SaveImage / PreviewImage.

The inputs that matter

• prompt - required; empty prompt is a hard ValueError. Describe the modification, not the whole scene. The model already sees the source images.
• image_url - the source images as URLs (string or array, max 10). This is where UploadImage comes in: take your local IMAGE, run it through WaveSpeed Upload Image, and wire its all_image_urls output into this input. The URLs it produces are short-lived, so keep this node and the upload in the same workflow run.
• max_images - 1–15, default 4. Output count, and every one is billed.
• size_preset / width / height - same resolution controls as the T2I node; preset or Custom, 1440–4096px.
• seed - the cache-control trick from the whole pack applies here too: randomize re-runs and re-bills every queue; a fixed seed lets ComfyUI serve the cached result and skip the call.

The client input is optional - with a key in ERPK Settings the node builds its own client, like every other WaveSpeed node.

Installing the pack

Identical to the rest of the ERPK Collection:

1. ComfyUI Manager → Install Custom Nodes → search erpk → ERPK Custom Nodes → install → restart.
2. Or:

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

Set the WaveSpeed API key in right-click canvas → ERPK Settings → API Keys, and you're done. No weights, no VRAM requirements - the edits happen on ByteDance's servers via WaveSpeed.

Gotchas

• ~$0.035 per output image per the pack docs - a 4-image edit runs about 14¢, a 15-image run about 52¢, and that's on top of whatever the source images cost to make. Batch output is where the bill sneaks up.
• The expired-URL trap. image_url values that sat around too long will fail. If you get an image error, re-run the upload node - don't dig through saved URLs.
• It's a paid API reseller. Pricing and Seedream behavior drift; the pack's README is the current source of truth, not a forum post from last quarter.
• Old workflows referencing "WaveSpeed Custom SeedreamV4EditSequential" migrate to this node type automatically, so pre-V3 saves keep loading.