Introduction

Seedream 5.5, ByteDance's newest image generation and editing model, is now available in ComfyUI through Partner Nodes. Comfy announced it on November 11 with a launch post on the Comfy blog and two ready-made templates, one for text-to-image and one for image edit. Both run in local ComfyUI or in Comfy Cloud. Seedream 5.5 is a closed, API-only model, so there are no weights to download. Runs are billed to your Comfy account in credits.

We covered the model itself when ByteDance unveiled it on November 9. What's new today is that you can drop it into a node graph: chain several edits in a row, then send the transparent results straight into your compositing nodes.

What Shipped

The code landed in ComfyUI PR #17631, merged on November 10. It adds seedream 5.5 to two existing ByteDance nodes:

  • ByteDance Seedream 4.5, 5.0 & 5.5 (ByteDanceSeedreamNodeV3) handles text-to-image and reference-based editing, with up to 10 reference images
  • ByteDance Seedream 5.0 & 5.5 Layer Separation (ByteDanceSeedreamLayerSeparationNodeV2) splits one image into a background plate plus up to 99 transparent layers with Seedream 5.5, up from 16 on 5.0. Each layer comes with its stacking order, bounding box, name, and description

Both nodes sit under partner/image/ByteDance in the node library.

The Knobs That Matter

  • Background: auto, opaque, or transparent. This control only appears when 5.5 is selected. Transparent works best when your prompt describes an isolated subject. With auto, the model decides and mostly returns transparency for stickers and logos. Output is always PNG, so the alpha channel survives
  • Size presets at 1K, 1.5K, and 2K in eight aspect ratios from 1:1 to 21:9, with 2K at 1:1 (2048×2048) as the default. You can also set a custom size between 0.92 MP and 4.62 MP, at any aspect ratio from 1:16 to 16:1
  • No batch and no prompt-thinking toggle on 5.5. You get one image per run, so use the seed when you're comparing variations
  • Layer Separation max_images defaults to 30 and goes up to 100, counting the base image. The model decides how many layers it actually needs, and every returned image is billed. crop_layers defaults to minimal size on 5.5, so each layer comes back trimmed to its bounding box

Pricing

These are the approximate per-image prices from the node's own pricing logic, billed in Comfy credits:

  • 1K or 1.5K output: about $0.0286 per image
  • 2K output: about $0.05005 per image
  • Each reference image: about $0.0114 extra
  • Layer Separation: the same 1K or 2K rate, multiplied by the number of layers returned

That makes a 2K edit with two references about $0.073. A layer split that returns 30 layers at 1K costs about $0.86, so set max_images deliberately.

How to Run It Today

  1. Update ComfyUI to the latest version, or open Comfy Cloud
  2. Sign in to your Comfy account, since Partner Nodes run on credits
  3. Open the Templates panel and load Seedream 5.5: Text to Image or Seedream 5.5: Image Edit
  4. Write a prompt, attach references if you're editing, pick a background mode, and run

Here's a starter prompt for the transparent mode. Set background to transparent and use a 1K 1:1 preset:

A die-cut vinyl sticker of a cheerful paper-lantern fox curled around a glowing crescent moon, thick clean white outline, soft cel-shaded lilac and aqua tones with warm gold highlights, isolated subject, no background

And a two-step edit chain to try with the Image Edit template:

Replace the sofa with a mid-century walnut daybed upholstered in deep teal velvet, keep the room layout, window light, rug, and every other object exactly the same

Now change only the wall art above the daybed to a large abstract gold-leaf painting, keep everything else identical

Why It Matters

  • Native alpha in a graph. Getting a transparent output used to mean generating, removing the background, and cleaning up halos. With Seedream 5.5 it's one node and a PNG, ready for a composite or a sticker sheet
  • Layer Separation at 99 layers. That's enough to break down a busy poster or a product shot into elements you can move around, which is real territory for designers working on layouts
  • Cheap enough to chain. At about 3 cents per 1K image, running several rounds of edits costs less than many single closed-model generations

Original Source

Conclusion

Seedream 5.5 is now available in the tool where the rest of your pipeline lives. Load the Image Edit template, push one photo through three rounds of targeted changes, and check whether the lighting and layout really hold. Then try Layer Separation on a poster you already have and watch it come apart into pieces you can move.

Try this on Gen → https://artrealmai.com/gen?utm_source=magazine&utm_campaign=gen&utm_content=seedream-5-5-comfyui-partner-nodes-rgba-layers

—Aurelia ♡