RunPod Seedream 4.0 Edit

Seedream 4.0 Edit is ByteDance's instruction-based editor, and the reason to reach for this specific node over its T2I sibling is the multi-image input: it accepts two image wires (input_image_1 and input_image_2) plus a text instruction. That's the "dress the model in the clothes from the second image" workflow - reference and subject fed together, one sentence, done. Closed model, no local option, so a RunPod public endpoint (seedream-v4-edit) is the way it gets into your graph at all.

Editing this way is exactly what the pack's README leans on for its image-editing nodes: optional image inputs, single or multiple, converted server-side. For Seedream 4.0 Edit that means the multi-image slots are the whole point - most edit nodes in this pack only take one input image, so this is where you go when the job needs a second reference.

How it works

Async POST to seedream-v4-edit, polling in the console, base64 image back to an IMAGE tensor. Both input_image_1 and input_image_2 are encoded as data URLs and collected into an images array in the payload - along with whatever you put in the images JSON text field, which gets merged with them. So you have two ways in: wire images from your graph, or paste URLs into the JSON. Wires are easier.

The inputs that matter

• input_image_1 / input_image_2 - the two sources. Subject in one, reference (clothes, background, style) in the other. Wire these from Load Image nodes.
• prompt - the instruction: "Dress the model in the clothes and hat" is the pack's default example, and it's a perfect illustration of the multi-image use case.
• images - a JSON array of image URLs, for when your references are already hosted. Only needed if you're not using the wires.
• size - string, default 1024*1024.
• enable_safety_checker - on by default, enforced at the source.

What comes out

One image output (IMAGE). Save it, preview it, chain an upscaler - standard tensor.

Installing it

One of 27 nodes in tcarwash/ComfyUI_RunpodNodes. Install via ComfyUI Manager (search ComfyUI_RunpodNodes) or:

cd ComfyUI/custom_nodes
git clone https://github.com/tcarwash/ComfyUI_RunpodNodes

Restart. No extra dependencies - the pack only needs what ComfyUI already ships. You need a RunPod key from runpod.io.

Gotchas

• Closed + metered - per-image billing, no weights to ever own, filtering at the source. The standard closed-model trade.
• Queue waits - 30–120 seconds; the console's [RunPod] Status: lines are your progress.
• "Job failed" - key or credits first, then check that your images JSON is valid if you used it.
• The two slots are the feature - if you only feed one image and a prompt, you're using half the node. Bring the second reference.

Multi-image editing is where closed editors currently show off - Nano Banana and Seedream 4.0 both lean on it - and this is the Seedream version, wired straight into ComfyUI. If your edit job has a reference image, this is the node in the pack for it.