SF WaveSpeed Seedream V4 Sequential

Storyboards, visual sequences, "here's the scene beat by beat" - that's a job most image generators are bad at, because each run forgets the last. SF WaveSpeed Seedream V4 Sequential is the node that doesn't: one prompt describes an ordered series of images, and the model generates them as a sequence, with the continuity baked in. Up to 15 frames in one call.

It's the Seedream V4 sibling to the pack's plain Seedream node - same ByteDance API line, same high-res presets, but the "sequential" mode trades a single hero image for a coherent run of them. If you've ever chained four generations together by hand hoping a character would stay consistent, this is the feature you've been improvising.

It's part of the comfyui-stillfront pack, under Stillfront/WaveSpeed.

How it works

The node calls WaveSpeed's bytedance/seedream-v4/sequential endpoint with your prompt and the max image count, then returns the whole batch. The key mechanic is in the prompt itself.

The inputs that matter

• prompt - the tooltip is emphatic here: you must clearly specify the number of images in the prompt. Something like "Show 4 frames: a knight leaving a castle, crossing a bridge, fighting a dragon, and standing victorious." The model reads the count from the prose, and it has to agree with the other setting below.
• max_images - 1 to 15, default 4. This is your safety cap, and it must align with the number stated in the prompt. Mismatch is the #1 way this node misbehaves - either the model stops early or the count drifts.
• size_preset - the same nine high-res presets as the plain Seedream node (2048×2048 default, up to 4096×4096).
• seed - fixed seeds give you the same sequence again.
• enable_sync_mode - default off, same async-default quirk as the plain Seedream node.

Output is output_images - a batch IMAGE holding the whole sequence on one wire. Split it with a batch-split node if you need individual frames.

Installing it

Pack install plus WaveSpeed key:

cd ComfyUI/custom_nodes
git clone https://github.com/razvanmatei-sf/comfyui-stillfront
cd comfyui-stillfront
pip install -r requirements.txt

Restart, then wire in an SF WaveSpeed Client node with your key (or config.ini / WAVESPEED_API_KEY).

Troubleshooting

• Got the wrong number of images. Your prompt and max_images disagree. Re-read the tooltip's rule: the count must be stated in the prompt and match the cap. Fix both, not one.
• Sequence loses coherence halfway through. Long prompts with vague beats let the model drift. Each beat needs a concrete, describable image - and 15 frames is a lot to hold together; 4–6 is the reliability sweet spot.
• Slow, and the console says "task submitted." Async default again. It's polling; give it time, and consider flipping enable_sync_mode on for the tighter path.
• Per-frame cost. Every generated frame is billed. 15 frames at 4096×4096 is a real line item - draft small and short, then commit.

The one caution: "sequential" means story beats, not true temporal animation - these are still separate stills. But for storyboards, thumbnail sequences, and multi-panel concepts, it's a surprising amount of structured output from a single prompt.