Seedream4Txt2Img(NYJY)
Seedream 4.0 text-to-image in ComfyUI — 2K output, zero VRAM
- Image
- Width
- Height
Seedream is ByteDance's flagship image model, and it is stubbornly API-only - the KB's ByteDance panel makes that point bluntly: the entire Seedream line is closed, full stop. So if you want Seedream 4.0's output, you have two choices: call the API, or don't have it. Seedream4Txt2Img is the call-the-API option with a ComfyUI face on it.
It's the flagship node of the Volcengine section of ComfyUI_NYJY (aidenli's pack), and the reason it earns that spot is resolution: Seedream 4.0 generates at native 2K, with 4K available. On your own GPU that's a very expensive proposition. Here it's a dropdown.
How it works. The node sends your prompt to the Volcano Engine Ark image endpoint and decodes the returned base64 straight into a standard ComfyUI IMAGE tensor. The model defaults to doubao-seedream-4-0-250828 (the 2025-08-28 build of 4.0). The field is free text, and the pack's own model list also includes the newer doubao-seedream-4-5-251128 - you can type it in if you want the latest revision.
The inputs that matter:
ratio_or_size- eleven options. Eight are aspect presets at native 2K (1:1 2048x2048,16:9 2560x1440,21:9 3024x1296, and so on), then three straight resolution buckets:1K,2K,4K. Pick an aspect ratio for composition control, or a K-bucket for raw resolution.override_with/override_height- set both above zero to force a custom pixel size; otherwise it uses the preset. This is your escape hatch for exact dimensions.max_images- 1 to 15. This is the sneaky-good feature: Seedream 4.0 supports sequential image generation, so ask for 4 images and you get a batch back from one prompt, useful for sampling variations in one call.prompt- your text-to-image prompt. Seedream is a strong prompt-follower, so write it like you mean it.seed,watermark,api_key- optional; seed for reproducibility, watermark off by default, api_key only if you're overriding the pack config.
The outputs. Image is the result - a batch tensor if max_images is above 1. Width and Height report the returned size of the first image, handy for downstream sizing.
The gotchas. The pack's shared failure mode applies: on API error the node returns a blank black 512×512 image and logs the real error to the console - black output means "go read the log". The API key is the same story as the rest of the pack: a Volcano Engine Ark key, and non-China sign-up has historically been the hard part. And remember this is pay-per-image - a batch of 15 at 4K is real money, so treat max_images like a budget control, not a toy.
Install. It ships with the pack:
cd ComfyUI/custom_nodes
git clone https://github.com/aidenli/ComfyUI_NYJY
pip install -r requirements.txt
or ComfyUI Manager → "Install via Git URL" → https://github.com/aidenli/ComfyUI_NYJY. The dependency that actually matters here is volcengine-python-sdk[ark].
When to reach for it. When you want current-generation Seedream quality without the hardware - or when 2K+ native output would melt your local setup. If you're already on Seedream 3.0, this is the upgrade path sitting in the same pack.
Inputs (9)
| Name | Type | Default | Description |
|---|---|---|---|
| model | STRING | doubao-seedream-4-0-250828 | — |
| prompt | STRING | — | |
| ratio_or_size | COMBO | 1:1 2048x2048 | 11 options: 1:1 2048x2048, 2:3 1664x2496, 3:4 1728x2304, 9:16 1440x2560, 3:2 2496x1664, 4:3 2304x1728, +5 |
| override_with | INT | 0 | — |
| override_height | INT | 0 | — |
| max_images | INT | 11–15 | — |
| api_keyopt | STRING | — | |
| watermarkopt | BOOLEAN | false | — |
| seedopt | INT | 00–2147483647 | — |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| Image | IMAGE | — |
| Width | INT | — |
| Height | INT | — |