SF WaveSpeed Seedream V4 Edit Sequential

Single-image editing is table stakes these days. Generating a consistent series - a character sheet, a turn-around, a pose set for the same subject - is the part that usually means writing a script, looping a generation node, and praying each pass doesn't drift. This node collapses that loop into one call: you describe the set, tell it how many images you want, and Seedream V4's edit-sequential endpoint returns the whole batch in one shot.

It's the sibling of the plain SF WaveSpeed Seedream V4 Edit node, from the SF ComfyUI Nodes pack (Stillfront/comfyui-sf-nodes). Where the single-edit node takes one image and one instruction, this one takes up to ten input images and an explicit count, and aims for consistency across the outputs. If you've ever needed "the same character from five angles" or "the same product in three scenes," this is the node purpose-built for that.

How it works

The mechanism is nearly identical to the plain edit node - POST to /api/v3/bytedance/seedream-v4/edit-sequential with your images, prompt, max_images count and size - but the API does the hard part: it generates max_images results designed to read as one set. The prompt does a lot of the consistency work, which is why the tooltip is unusually prescriptive: specify the number of images in the prompt and use phrases like "a series of" or "group of images." Say "a series of 4 images of this character in different poses," not just "make four images."

The output comes back as output_images, a batched IMAGE tensor - one item per generated image - so it feeds straight into a batch-splitting or grid-montage node.

The inputs that matter

• client - required, from the SF WaveSpeed Client node.
• prompt - describe the edit/generation task and the set. This is the field the author tells you to align with max_images; they need to agree.
• max_images - 1 to 15, default 2. How many images to generate. Should match the count you put in the prompt.
• size_preset - the same nine Seedream V4 presets as the single-edit node, up to 4096x4096. Note that 15 images at 4K is a lot of billable generation; size down for drafts.
• seed - 0 means random; set it to lock the series and iterate on the prompt.
• image_1 … image_10 - optional input images as URL strings (connect from ComfyUI's Upload Image node). You can run this with zero inputs - the tooltip allows it as an editing/generation task - but at least one input is enforced in code, and honestly the reference-image path is where it shines.
• enable_sync_mode - default off; the node polls until the set is done either way.

Install

Same pack, same story. ComfyUI Manager → search "SF ComfyUI Nodes" → install → restart, or:

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

Then connect an SF WaveSpeed Client node with your WaveSpeed key (paste it in, or use config.ini [API] api_key / WAVESPEED_API_KEY). No models to download.

Gotchas

• The prompt/count mismatch is the #1 failure. Set max_images to 4 but describe three images and you'll get unpredictable output. State the count in the prompt.
• Watch the bill. max_images multiplies the per-image cost. The tooltip's own defaults are modest; 15 images at the biggest preset is a real charge.
• Image URLs must be publicly reachable - WaveSpeed's servers fetch them, so no localhost, no local file paths.
• The README's wavespeed_api_key config example doesn't match what the code reads (api_key); copy config.ini.tmp.

The honest comparison: for the same "consistent series" job locally you'd be chaining Qwen Edit 2511 calls and hoping faces hold up. This is the node that does it server-side in one shot - at a price, but with none of the drift babysitting. For character-sheet work specifically, it's the easiest set-generator in the pack.