EmpirioLabs AI media backends for Hermes Agent

EmpirioLabs AI is a specialized AI inference and integration provider. It hosts open-source models on GPU infrastructure it manages, runs optimized endpoints for proprietary models, helps teams ship their own models, and offers on-demand GPU Cloud instances and hosted AI agents. Its model catalog spans text, image, audio, video, transcription, embedding, reranking, research, and 3D models.

This plugin adds three EmpirioLabs AI backends to Hermes Agent, each named empiriolabs:

Hermes tool	Backend setting	What it does
`image_generate`	`image_gen.provider: empiriolabs`	Image generation and image editing
`video_generate`	`video_gen.provider: empiriolabs`	Text-to-video and image-to-video
`text_to_speech`	`tts.provider: empiriolabs`	Text-to-speech

It registers no tools, hooks or commands of its own. For chat models, install the companion plugin empiriolabs.

Requirements

Hermes Agent 0.21.6 or later
An EmpirioLabs AI API key from https://platform.empiriolabs.ai/dashboard/api-keys

Install

From the Hermes plugin catalog:

hermes plugins install empiriolabs-media

From GitHub (add --ref <full commit SHA> to pin a reviewed commit):

hermes plugins install EmpirioLabs-AI/hermes-empiriolabs/media

Enable it when the installer asks, or later with hermes plugins enable empiriolabs-media. The installer also asks for EMPIRIOLABS_API_KEY and saves it to your Hermes .env file.

Choose models

Run hermes tools, open Image Generation or Video Generation, select EmpirioLabs AI and pick a model; the lists come from the live catalog at GET https://api.empiriolabs.ai/v1/models. For Text-to-Speech, select EmpirioLabs AI in the same menu and set a model and voice in config.yaml when you want something other than the defaults.

You can also set everything in config.yaml:

image_gen:
  provider: empiriolabs
  model: flux-2-klein-4b
video_gen:
  provider: empiriolabs
  model: hunyuan-video-1-5
  empiriolabs:
    timeout_seconds: 900   # how long one video_generate call waits
tts:
  provider: empiriolabs
  model: tts-2-flash
  voice: Sarah

Without a model setting, each backend uses the model named above when it is in the catalog, and otherwise the first featured model of that kind. A model id that is not in the EmpirioLabs AI catalog (for example one left over from another backend) is ignored. image_gen.empiriolabs.model, video_gen.empiriolabs.model, tts.empiriolabs.model and tts.empiriolabs.voice set a model or voice for this backend only.

How each backend works

Images. One request to POST /v1/images/generations. When a generation runs long, the plugin polls GET /v1/jobs/{job_id} for up to 10 minutes. To edit an image, the agent passes image_url: an https URL, a data:image URI, or a local PNG, JPEG, WebP or GIF file of up to 20 MB, which is sent inline. Editing is offered when the selected model accepts input images.
Videos. One request to POST /v1/videos/generations, then GET /v1/jobs/{job_id} every 5 seconds for up to 15 minutes (video_gen.empiriolabs.timeout_seconds). If the wait ends first, the job keeps running and the tool result gives its job id and the URL to fetch it from. Duration, resolution and aspect ratio are matched to the values the model offers. Pass image_url to animate an image.
Speech. One request to POST /v1/audio/speech per piece of text. Voices and output formats come from the selected model. When the text is longer than the model accepts in one request, it is split at sentence boundaries and the MP3 pieces are joined.
Generated images and videos are downloaded into Hermes' cache ($HERMES_HOME/cache/generated/); if that download fails, the tool result gives the signed URL instead. Speech is written to the file Hermes asks for.

Data handling

Prompts, source images and the text to speak go to api.empiriolabs.ai (or EMPIRIOLABS_BASE_URL) with your EMPIRIOLABS_API_KEY. Each image, video and speech request is billed to the EmpirioLabs AI account that owns the key, and Hermes does not ask for approval before a backend call.
Generated files are downloaded from the URLs the API returns. The key is never sent to those URLs, and every download address is checked with Hermes' URL safety check.
To edit a local image, the plugin reads that one file. It reads no other files, writes only the generated media into Hermes' cache, starts no background processes, runs no shell commands and sends no telemetry. It reads your key through Hermes and never stores it.
EmpirioLabs AI privacy policy: https://empiriolabs.ai/privacy-policy

License

MIT

empiriolabs-media

What it adds

Environment variables it needs 1

README