Open WebUI
Chat with ModelSite models and generate Seedream images in Open WebUI: connection settings and the image size you must change
Status: run against ModelSite with Open WebUI 0.12.3 on 2026-10-24: the model list loaded, a chat answered, and a Seedream image was generated.
Connect
Set these when you start Open WebUI (Docker -e, or your environment):
OPENAI_API_BASE_URL=https://api.modelsite.ai/v1
OPENAI_API_KEY=$MODELSITE_API_KEYOr, in the app: Admin Panel → Settings → Connections → OpenAI API, set the URL to https://api.modelsite.ai/v1 and paste the key.
Open WebUI saves its settings on first start. After that, changing these environment variables has no effect; change them in the admin panel instead.
On connect, Open WebUI lists every model ModelSite returns from /v1/models (38 in our test). The list includes video and image models next to the chat models, because the list does not mark a model's type. Pick a chat model such as glm-5.3-flash for conversations.
Image generation
In Admin Panel → Settings → Images:
| Field | Value |
|---|---|
| Image generation engine | OpenAI |
| API base URL | https://api.modelsite.ai/v1 |
| API key | your key |
| Model | doubao-seedream-4.5 (or doubao-seedream-5.0-lite, doubao-seedream-4.0) |
| Image size | 2048x2048 |
Open WebUI's default image size is 512x512, which is below Seedream's minimum, so the first request returns 400. The error message states the minimum pixel count and an example size (for instance 1920x1920); set that size in Open WebUI's image settings.
| Model | Smallest accepted size |
|---|---|
doubao-seedream-4.0 | 921,600 pixels (1024x1024 works) |
doubao-seedream-4.5, doubao-seedream-5.0-lite | 3,686,400 pixels (about 1920x1920) |
2048x2048 satisfies all three. In our run it returned a 2048×2048 image in about 10 seconds. Images are returned as files that Open WebUI stores itself; the temporary image link the API gives is valid for 24 hours.
Remove the "AI generated" badge
Seedream adds an "AI generated" badge to images by default. Open WebUI can pass extra parameters to the image API through its IMAGES_OPENAI_API_PARAMS setting. Set it to:
{"watermark": false}(as an environment variable on first start, or in the same setting afterwards). In our run the next image came back without the badge.
Limits
- The model list mixes video and image models with chat models.
- Video generation is not part of Open WebUI; for video use the n8n HTTP recipe or the Video SDKs.