ModelSiteAPI

LobeChat

Use ModelSite in LobeChat through the Qwen provider for chat and Wan video, and why the Volcengine provider does not work

Status: read from source (LobeChat 2.2.7, commit e93e659). We read the provider code to see which paths it calls; we did not run LobeChat against ModelSite.

LobeChat configures a base URL per provider, and its video generation depends on the provider.

Qwen provider: works for chat and Wan video

Set the provider's API proxy URL to https://api.modelsite.ai/compatible-mode/v1 and enter your key.

  • Chat and the model list: LobeChat calls {base}/chat/completions and {base}/models; ModelSite serves both under /compatible-mode/v1.
  • Wan video: LobeChat removes the /compatible-mode/v1 suffix and calls /api/v1/services/aigc/video-generation/video-synthesis to create the task and /api/v1/tasks/{id} to poll it, with Authorization: Bearer. These are ModelSite's Wan entrances.

Volcengine provider: not supported

LobeChat's Volcengine provider creates Seedance tasks with a callback_url and then waits for a callback instead of polling. ModelSite rejects callback_url with 400 and does not send callbacks, so video generation through this provider does not work. Use the Ark SDK or a plain HTTP client to poll Seedance tasks.

Limits

  • Image generation in LobeChat was not checked.
  • Because this is a read of the source, confirm with one short chat message after you set the base URL.

new-api

Add ModelSite as an upstream channel in new-api so your own users can call Wan, HappyHorse and Seedance through your relay

Models

GET /v1/models and GET /v1/models/{id}: query the models you can call, and how model IDs are used

On this page

Qwen provider: works for chat and Wan videoVolcengine provider: not supportedLimits