WaveSpeedAI ByteDance Seedream V4 Sequential

Some prompts aren't a single image - they're a set. "Four poses of the same character," "a 12-frame product rotation," "three colorways of this sneaker." That's exactly what this node is for: one prompt in, a batch of Seedream V4 images out, all in a single output_images tensor. It's the text-to-image version of the Seedream V4 Sequential family from the razvanmatei-sf/razv-wavespeed pack, which means it's a paid call into the WaveSpeed AI cloud - no local weights, no GPU, and no free lunch.

If you've ever generated a series one image at a time and watched the style wander, this is the appeal in one sentence: the model holds the whole set in mind while it renders, so the batch stays stylistically consistent in a way that stitching together 15 independent generations never does.

How it works

Type a prompt, set max_images, and the node POSTs to the Seedream V4 sequential endpoint. Async mode is the default - it grabs a task ID and polls until the whole batch is done - then converts the returned URLs into one batched IMAGE tensor. That output is batch-shaped, so feed it to batch-aware preview/save nodes and you'll get the set out in one go.

Same rule as its edit sibling, and the tooltip is blunt about it: your prompt must clearly specify the number of images, and max_images has to line up with what you described. "Generate 5 variations" with max_images set to 4 is a contradiction the model will resolve in a way you won't like.

The inputs that matter

• prompt - name the count, describe the set. "A series of 6 illustrations of a cat, each in a different art style" beats "cat, 6 images" every time.
• max_images - 1–15, default 4. Match it to the prompt.
• size_preset - the standard nine Seedream presets, default 2048x2048 (1:1), up to 4096×4096.
• seed - Seedream quirk again: 0 = random. Set a real number to reproduce the set.

enable_sync_mode is the only other thing on the panel; it defaults off, and async is the right call for a batch anyway.

Install

Shared pack:

cd ComfyUI/custom_nodes
git clone https://github.com/razvanmatei-sf/razv-wavespeed
# restart ComfyUI

Or ComfyUI Manager → search "razv-wavespeed" → Install. Then a WaveSpeed AI key: the WaveSpeedAI Client node takes it directly, or copy config.ini.tmp → config.ini in the pack folder ([API] section), or export WAVESPEED_API_KEY.

Common issues

• Count mismatch is the recurring failure mode here and in the edit variant. Prompt says N, max_images says N, or it's a gamble.
• Batch size × resolution = bill. 15 images at 4096×4096 is the most expensive single click in this pack. Presets are de facto price tiers.
• Big batches can take minutes - it's one task, not 15, but it's the slowest node in the family. Async mode exists for exactly this.
• No upload involved (it's text-to-image), so the "Upload Image node" confusion that hits the edit nodes doesn't apply here. Just the key, the prompt, and the count.