hermes-openrouter-image-gen

OpenRouter image generation plugin for Hermes Agent.

What it does

Adds OpenRouter as an image generation backend via /v1/chat/completions — for image models (Gemini, GPT Image, Grok, FLUX, Seedream) that return images inline in chat responses.

Features:

• Text-to-image via 8 models (Gemini Flash/Pro, GPT-5 Image Mini, Grok Imagine, Seedream 4.5, FLUX.2 Pro)
• Multi-image input for merge/edit/compose via image_url + reference_image_urls
• Image-to-image editing (primary source + reference images)
• Native aspect ratio control via image_config API parameter
• Safety settings: BLOCK_NONE for all 4 categories
• Data URI + URL + local file path input handling
• Response parsing: handles snake_case, camelCase, content parts, bare URLs, raw base64

Models

Model	Speed	Price	Best For
Nano Banana 2 (Flash)	~3-8s	~$0.07	Fast, good quality
Nano Banana 2 (Flash) Stable	~3-8s	~$0.07	Same, stable endpoint
Nano Banana Pro	~8-15s	~$0.15	Best quality, 2K/4K
Nano Banana Pro Stable	~8-15s	~$0.15	Same, stable endpoint
Grok Imagine	~5-10s	~$0.012/K	Photorealism, text
Seedream 4.5	~5-10s	$0.04 flat	Editing, portraits
FLUX.2 Pro	~6s	~$0.03/MP	Studio quality
GPT-5 Image Mini	~10-15s	~$0.008	Cheapest option

Installation

# Clone
git clone https://git.ashu.io/ashuio/hermes-openrouter-image-gen.git /tmp/or-img

# Deploy plugin
mkdir -p ~/.hermes/plugins/image_gen/openrouter
cp /tmp/or-img/plugin/* ~/.hermes/plugins/image_gen/openrouter/

# Enable
hermes config set image_gen.provider openrouter
hermes config set image_gen.model google/gemini-3.1-flash-image-preview

Docker deployment

FROM nousresearch/hermes-agent:latest

# Clone plugin
RUN git clone --depth 1 https://git.ashu.io/ashuio/hermes-openrouter-image-gen.git /tmp/or-img && \
    mkdir -p /opt/data/plugins/image_gen/openrouter && \
    cp /tmp/or-img/plugin/* /opt/data/plugins/image_gen/openrouter/ && \
    rm -rf /tmp/or-img

Config

image_gen:
  provider: openrouter
  model: google/gemini-3.1-flash-image-preview
  openrouter:
    model: google/gemini-3.1-flash-image-preview

plugins:
  enabled:
    - image_gen/openrouter

Multi-image usage

# Text-to-image
image_generate(prompt="a sunset over mountains", aspect_ratio="landscape")

# Image editing (new ABC contract)
image_generate(
    prompt="make this a watercolor painting",
    image_url="/path/to/photo.jpg",
)

# Multi-image merge (new ABC contract)
image_generate(
    prompt="swap the legs between these two women",
    image_url="/path/to/image1.jpg",
    reference_image_urls=["/path/to/image2.jpg"],
)

# Legacy format (still supported)
image_generate(
    prompt="merge these images",
    images=["/path/to/image1.jpg", "/path/to/image2.jpg"],
)

Upstream compatibility

v2.0 (current): Matches the upstream ImageGenProvider ABC contract:

• generate() accepts image_url and reference_image_urls as keyword-only params
• capabilities() declares modalities: ["text", "image"] and max_reference_images: 9
• Uses normalize_reference_images() helper
• Uses image_config for native aspect ratio control (replaces prompt injection)
• Uses modalities: ["image", "text"] in API payload
• Handles both snake_case and camelCase response formats

No upstream tool patch required. The upstream image_generation_tool.py already passes images= to the plugin dispatch. The patches/ directory is retained for legacy Hermes versions only.

Cost

~$0.07/image for Gemini 3.1 Flash, ~$0.15/image for Gemini Pro, $0.04 flat for Seedream.