Apikey Image Gen

Studio Image Generation

Use this skill when the user wants to generate an image or edit an existing image.

Always call Ekko Studio's media endpoint. Do not call an upstream image API directly, and do not ask the user for an API key. The server reads the selected/requested profile's config.yaml and uses a configured custom provider. By default it uses the provider named fun-codex, but callers may request another configured provider by sending provider, provider_name, or custom_provider.

This skill is separate from Hermes Agent's native image_generate tool. The native tool reads image_gen from config.yaml; this Studio-managed endpoint reads the auxiliary.image_generation and auxiliary.image_edit routes below. Changing one does not change the other.

Do not use any built-in image generation tool as a fallback. If the Hermes Web UI endpoint returns 401, 403, connection failure, or any other error, stop and report the Hermes Web UI error to the user.

custom_providers:
  - name: fun-codex
    base_url: https://api.apikey.fan/v1
    api_key: ...
    model: gpt-5.5
    api_mode: codex_responses

Example with another configured provider:

custom_providers:
  - name: agnes
    base_url: https://agnes.example/v1
    api_key_env: AGNES_API_KEY
    model: agnes-image-2.1-flash

auxiliary:
  image_generation:
    provider: agnes
    model: gpt-image-2
    timeout: 600
  image_edit:
    provider: agnes
    model: gpt-5.4-mini
    timeout: 600

image_generation supplies the route for text-to-image and multipart image edits. image_edit supplies the route and primary model for image-to-image through the Responses API. If the active route has no provider, the server falls back to the other image route and then to fun-codex.

Endpoint:

POST <Hermes Web UI base URL>/api/studio/media/apikey-image-generate

Resolve the Hermes Web UI base URL in this order:

1. HERMES_WEB_UI_URL environment variable, if set.
2. http://127.0.0.1:${PORT}, if PORT is set.
3. http://127.0.0.1:8648 for the Web UI single-server default.

Common local ports:

• Development API backend: http://127.0.0.1:8647. Use this with npm run dev; do not target the Vite frontend port.
• Web UI single-server default: http://127.0.0.1:8648.
• Desktop app default: http://127.0.0.1:8748.
• Custom port: set HERMES_WEB_UI_URL to the full base URL, or set PORT to use http://127.0.0.1:${PORT}.

When Hermes Web UI is running from Docker Compose, the default external URL is http://127.0.0.1:6060.

Authentication:

Send the Hermes Web UI server bearer token. This token is accepted only by Hermes Web UI media generation endpoints for agent skills; it is not a general Web UI login token.

Resolve the token in this order:

1. AUTH_TOKEN environment variable, if set.
2. ${HERMES_WEB_UI_HOME}/.token, if HERMES_WEB_UI_HOME is set.
3. ${HERMES_WEBUI_STATE_DIR}/.token, if HERMES_WEBUI_STATE_DIR is set.
4. ~/.hermes-web-ui/.token.

Profile selection:

Use the current Hermes profile from the run instructions by sending X-Hermes-Profile.

If the run instructions include [Current Hermes profile: <name>], include:

-H "X-Hermes-Profile: <name>"

Replace <name> with the exact profile name from the run instructions. Never send a placeholder value such as <name> or <current-hermes-profile>.

If no current profile is provided, omit the header and let the server fall back to the current Hermes active profile.

Modes

Text To Image

Use when there is no input image.

{
  "mode": "text",
  "prompt": "A high quality product image of a matte black mechanical keyboard on a clean desk",
  "size": "1024x1024",
  "output_path": "/absolute/path/to/output.png"
}

The server calls POST /v1/images/generations against the configured auxiliary.image_generation provider, then falls back to fun-codex. If provider, provider_name, or custom_provider is present, the server calls the requested provider's base URL instead.

Image To Image

Use when the user provides an existing image and wants the model to modify or redraw it.

{
  "mode": "image",
  "prompt": "Use this reference composition and generate a refined technology brand poster",
  "image_path": "/absolute/path/to/reference.png",
  "size": "1024x1024",
  "output_path": "/absolute/path/to/output.png"
}

The server calls POST /v1/responses against the configured auxiliary.image_edit provider, then the image-generation provider, then fun-codex. If provider, provider_name, or custom_provider is present, the server calls the requested provider's base URL instead.

Image Edit

Use when the user wants to modify an existing image while preserving parts of it.

{
  "mode": "edit",
  "prompt": "Change the background to blue and keep the subject unchanged",
  "image_path": "/absolute/path/to/source.png",
  "size": "1024x1024",
  "output_path": "/absolute/path/to/edited.png"
}

The server calls POST /v1/images/edits against the configured auxiliary.image_generation provider, then falls back to fun-codex. If provider, provider_name, or custom_provider is present, the server calls the requested provider's base URL instead.

Request Fields

• mode: text, image, or edit.
• prompt: required.
• provider: optional configured custom provider name. Defaults to fun-codex. custom:<name> is accepted and normalized to <name>.
• provider_name: optional alias for provider.
• custom_provider: optional alias for provider.
• image_path: local png, jpeg, or webp path. Required for image and edit unless using image_url or image_base64.
• image_url: optional alternative image input.
• image_base64: optional alternative image input. If it is not a data URI, include mime_type.
• n: number of images. Defaults to 1.
• size: defaults to 1024x1024. Common values: 1024x1024, 1536x1024, 1024x1536, 2048x2048, 3840x2160, 2160x3840, auto.
• quality: defaults to auto.
• model: optional override. If omitted, text/edit modes use auxiliary.image_generation.model, then gpt-image-2; image mode uses auxiliary.image_edit.model, then the selected provider's model, then gpt-5.4-mini.
• image_model: optional image tool model for image mode. If omitted, uses auxiliary.image_generation.model, then gpt-image-2.
• output_path: optional absolute output file path. If omitted, the server saves to ${HERMES_WEB_UI_HOME:-~/.hermes-web-ui}/media/*.png.
• timeout_ms: overrides the active auxiliary route's timeout (seconds). If neither is configured, defaults to 600000 milliseconds.

Curl Template

TOKEN="${AUTH_TOKEN:-}"
if [ -z "$TOKEN" ] && [ -n "${HERMES_WEB_UI_HOME:-}" ] && [ -f "$HERMES_WEB_UI_HOME/.token" ]; then
  TOKEN="$(cat "$HERMES_WEB_UI_HOME/.token")"
fi
if [ -z "$TOKEN" ] && [ -n "${HERMES_WEBUI_STATE_DIR:-}" ] && [ -f "$HERMES_WEBUI_STATE_DIR/.token" ]; then
  TOKEN="$(cat "$HERMES_WEBUI_STATE_DIR/.token")"
fi
if [ -z "$TOKEN" ] && [ -f "$HOME/.hermes-web-ui/.token" ]; then
  TOKEN="$(cat "$HOME/.hermes-web-ui/.token")"
fi
if [ -z "$TOKEN" ]; then
  echo "Missing Hermes Web UI token. Check AUTH_TOKEN, HERMES_WEB_UI_HOME, HERMES_WEBUI_STATE_DIR, or ~/.hermes-web-ui/.token." >&2
  exit 1
fi

BASE_URL="${HERMES_WEB_UI_URL:-}"
if [ -z "$BASE_URL" ]; then
  BASE_URL="http://127.0.0.1:${PORT:-8648}"
fi
BASE_URL="${BASE_URL%/}"

curl -sS -X POST "$BASE_URL/api/studio/media/apikey-image-generate" \
  -H "Authorization: Bearer $TOKEN" \
  -H 'Content-Type: application/json' \
  -d '{
    "mode": "text",
    "provider": "fun-codex",
    "prompt": "A cinematic 4K photo of a silver robot hand holding a small glowing cube",
    "size": "3840x2160",
    "output_path": "/absolute/path/to/output.png"
  }'

Successful responses include:

{
  "ok": true,
  "mode": "text",
  "output_paths": ["/absolute/path/to/output.png"],
  "provider": "fun-codex",
  "base_url": "https://api.apikey.fan/v1"
}

If the response code is missing_fun_codex_provider, tell the user to configure fun-codex in the selected/requested profile's config.yaml. If the response code is missing_apikey_image_provider, tell the user to configure the requested provider in the selected/requested profile's config.yaml, or omit provider to use the default fun-codex provider.