Latest news
AI ModelsSeedream 5.5ComfyUIByteDance

Seedream 5.5 Announced: 100-Step Editing and Transparent PNGs

ByteDance’s Seed team announced Seedream 5.5 on November 9. The model offers iterative image editing and transparent PNG output, and has been available through a paid API node in ComfyUI since November 10.

Contemporary interior visualization featuring decorative elements on a transparent background
© seedream5.one

In brief

  1. Seedream 5.5 was announced with iterative editing designed to preserve areas that are not being changed.
  2. The model can directly generate images with fully or partially transparent backgrounds.
  3. ComfyUI v0.45.0 added a paid API node that calls Seedream 5.5.
  4. The estimated cost shown in the ComfyUI node is about $0.029 per image for 1K and 1.5K outputs.

ByteDance’s research team, Seed, announced its image-generation model Seedream 5.5 on November 9. Its key features are an aim to preserve unchanged areas while making successive edits to the same image, and the ability to directly create images with alpha channels and transparent backgrounds. Seedream 5.5 also became available through a partner node that uses an external API with the ComfyUI v0.45.0 update released on November 10.

Iterative editing and transparent output

According to information shared by ByteDance, Seedream 5.5 focuses on preserving the appearance and colors of areas that are not being edited. The company’s examples show a single portrait being edited 100 times in succession, while small details such as a ring or eyeglass frames are preserved. This is a demo provided by the developer; independent test conditions and results have not been shared.

The model can also create partially transparent backgrounds as well as fully transparent ones. ByteDance says this could reduce the need to remove backgrounds and clean up edges afterward. The company also reports more natural results for lighting and material textures, and less distortion in small or dense text.

Using Seedream 5.5 through ComfyUI and pricing

Seedream 5.5 can be selected as the model in the “ByteDance Seedream 4.5, 5.0 & 5.5” node in ComfyUI. Image size can be set to 1K, 1.5K or 2K, or custom width and height values can be specified. The node accepts up to 10 reference images, and the background setting can be set to “auto”, “opaque” or “transparent”. The ability to generate multiple related images in a single run is not available in 5.5.

According to estimates in the node’s source code, 1K and 1.5K generations cost about $0.029, while 2K generations cost about $0.050; each reference image adds about $0.011. These are estimates shown in the node, not definitive billing amounts. Check the pricing of the service you use for the actual cost. The model is offered as a paid node that makes API calls to ByteDance’s servers, not as a package of weights that runs locally on a GPU in ComfyUI.

A layer-separation node included in the same source also supports Seedream 5.5. According to the node description, it can separate a single image into transparent layers whose backgrounds and positions can be changed; the upper limit for 5.5 is 99 layers. The input image must be at least 512 × 512 pixels.

Evaluations and availability

In ByteDance’s own evaluations, Seedream 5.5 is compared with Seedream 5.0 Pro across six use cases. The company reports that the “ready-to-use” rate is more than double that of the previous version in the film and television and office work categories. Expert comparisons claim a 55.62 percent improvement in the realism of image-to-image generation. Since the evaluation sample size and number of participants have not been disclosed, these results should be treated as the company’s own measurements.

Links to BytePlus AI Playground and Dola are listed for the model, and a phased rollout is also mentioned for some services. Pricing and available regions have not been disclosed on the official blog or product page. A more advanced Pro version is said to be coming later, but no date has been given.

Impact on architecture and visualization workflows

For architects, interior designers and archviz teams, transparent output could make it easier to isolate furniture, lighting or decorative elements for use in presentation images. Iterative editing is intended to let users change individual elements in a room image while preserving the rest. However, how consistently these features perform in real projects needs to be tested on sample scenes.

The ComfyUI integration makes it possible to add a node to an existing workflow, but generation takes place through an external API and is paid. As a result, API costs, service terms and how images are processed matter more than local GPU power. The sources do not say that the model produces 3D geometry or CAD output; its role here is limited to image generation and editing.

Sources

1 source
N
note.comnote.com/ai_image_goat/n/n302ed15e88d1
Summary

Source texts are not republished; short quotes are marked, everything else is our own summary and commentary.

3dsınıfı’s take
3dEditor’s assessment

Seedream 5.5’s most tangible promise for architecture and interior design teams is that it could reduce background cleanup when preparing visual assets and make it easier to apply staged changes to a single composition. Teams looking to quickly create alternatives for presentation images may find it worth considering; however, the 100-edit demo cited by the source does not prove that the same level of consistency will be achieved in every scene.

For offices in Turkey, decisions should take into account not only image quality but also paid API access and the total cost of generation. Prices shown in ComfyUI are estimates; check the actual price with the service provider. It is also worth keeping in mind that the model does not run locally and is not stated to offer 3D model generation.

Frequently asked questions

Can Seedream 5.5 be used in ComfyUI?

Yes. ComfyUI v0.45.0 added Seedream 5.5 as a paid partner node that calls an external API.

Can Seedream 5.5 generate transparent PNGs?

Yes. The model can create images with an alpha channel and fully or partially transparent backgrounds.

Does Seedream 5.5 run on a local GPU?

According to the source, ComfyUI support is provided as a node that makes API calls to ByteDance’s servers; the model does not run on a local GPU.

Comments and the forum are in Turkish.Join the discussion
+

Related news