Hermes Image Studio

AI image generation plugin for Hermes Agent. 11 style presets across 9 FLUX, GPT Image, and Nano Banana models. Negative prompts, social-ready sizes, prompt library, web gallery, cost tracking, inpainting, image-to-video, auto-upscaling, and editing, all through FAL.ai.

Generate publish-ready images for social media, articles, and creative projects directly from your Hermes conversations. No running to another tool.

Features

• 8 FLUX Presets — Cinematic, photorealistic, vintage, fantasy, illustration, minimalist, noir, and studio for FLUX models. Each preset tunes prompt phrasing, inference steps, and model selection for consistent visual results.
• 2 GPT Presets — gpt-photo for text rendering and complex multi-subject scenes (GPT Image 1.5), and gpt-art for creative and stylized interpretation (GPT Image 2).
• 1 Nano Banana Preset — banana-photo for photorealistic detail, accurate text rendering, and natural-language edits (Nano Banana 2).
• 9 Models — FLUX Pro, FLUX 2 Klein 9B, FLUX 2 Pro, GPT Image 1.5, GPT Image 2, Nano Banana 2, Nano Banana Pro, Clarity Upscaler, and the legacy FLUX Klein v1 for compatibility.
• Negative Prompts — Tell the model what to avoid: "no people, no text, no watermark."
• Social-Ready Sizes — Generate directly at Twitter post/header, Instagram story/post, and YouTube thumbnail dimensions.
• Auto-Upscaling — Every image can be 2x upscaled with a single command.
• Image-to-Image Editing — Transform any generated image with a new prompt using GPT Image 1.5. "Turn this photo into a watercolor painting" or "Add a dirt path leading to the door."
• Inpainting — Replace only the masked region of an image while preserving the rest (FLUX Pro Inpaint). Remove objects, change backgrounds, add elements precisely.
• Image-to-Video — Animate any still into a short MP4 clip with Kling v3 Pro or Veo 3.1 Fast. Camera moves, motion, and action from a text prompt.
• Prompt Library — Save your best prompts by name and regenerate them anytime.
• Web Gallery — Generate a self-contained HTML gallery of everything you've created, filterable by preset and model.
• Cost Tracking — Every generation records its estimated FAL cost. See spend per batch and in total.
• Batch Generation — Generate 2-8 variants simultaneously with random seeds to explore compositions and find the best result.
• History Tracking — Every generation is recorded in a local SQLite database. Browse, re-find, or reference past creations.
• Organized Outputs — Images are saved with descriptive filenames (20260726_103042_cinematic_472444619_cowboys-at-sunrise.png) to a configurable output directory.

Requirements

• Hermes Agent (any recent version)
• Python 3.10+
• A FAL.ai account with credits

Installation

Option A: Quick install (clone + symlink)

git clone https://github.com/YOUR_USERNAME/hermes-image-studio.git
ln -s "$PWD/hermes-image-studio" ~/.hermes/plugins/image-studio

Option B: Manual install

Copy the hermes-image-studio directory into ~/.hermes/plugins/:

cp -r hermes-image-studio ~/.hermes/plugins/image-studio

Enable the plugin

hermes plugins enable image-studio

Restart Hermes (or start a new session). The plugin auto-loads when FAL_KEY is set.

Setup

Get your FAL.ai API key:

1. Sign up at fal.ai
2. Go to fal.ai/dashboard and create an API key
3. Add it to ~/.hermes/.env:

FAL_KEY=your-api-key-here

That's it. Start a new Hermes session and you're ready.

Tools

The plugin registers 13 tools into the image-studio toolset:

Tool	What it does
image_studio_generate	Generate an image with a style preset, negative prompt, and social size
image_studio_edit	Edit an existing image with a new prompt (img2img)
image_studio_inpaint	Region-edit an image: replace only the masked area (FLUX Pro Inpaint)
image_studio_animate	Turn a still image into a short video clip (Kling v3 / Veo 3.1)
image_studio_upscale	2x upscale any generated image
image_studio_batch	Generate N variants with random seeds
image_studio_presets	List all available presets with descriptions
image_studio_gallery	Generate a self-contained HTML gallery of all your images
image_studio_save_prompt	Save a prompt to the library by name
image_studio_load_prompt	Generate from a saved prompt by name
image_studio_prompts	List the prompt library
image_studio_delete_prompt	Remove a saved prompt
image_studio_history	Browse recent generations with cost tracking

Quick Examples

# Generate a cinematic landscape
image_studio_generate(prompt="A lone cowboy riding through Monument Valley at sunrise, warm golden light, dust kicking up", preset="cinematic", aspect_ratio="landscape")

# Generate with a negative prompt (FLUX models)
image_studio_generate(prompt="A sunset over the ocean", negative_prompt="no people, no boats, no text")

# Generate at Twitter-post size (1200x675)
image_studio_generate(prompt="A product shot for a launch tweet", preset="studio", aspect_ratio="twitter-post")

# Edit that image (transform it with a new prompt)
image_studio_edit(image_url="https://...from-above-generation...", prompt="Add a herd of wild horses in the distance, thunderstorm brewing on the horizon")

# Region-edit: replace only the masked area (mask = white where to regenerate)
image_studio_inpaint(image_url="https://...", mask_url="https://.../mask.png", prompt="a red sports car parked by the cabin")

# Animate a still into a short video
image_studio_animate(image_url="https://...from-above-generation...", prompt="slow dolly-in, clouds drifting, golden light shifting", model="kling-video", duration="5")

# Save a prompt for reuse
image_studio_save_prompt(name="desert-sunrise", prompt="A lone cowboy riding through Monument Valley at sunrise", preset="vintage")

# Generate from a saved prompt
image_studio_load_prompt(name="desert-sunrise")

# List saved prompts
image_studio_prompts

# Batch 4 variants
image_studio_batch(prompt="A futuristic cityscape at night with neon lights", preset="fantasy", count=4)

# Upscale the result
image_studio_upscale(image_url="https://...")

# Build the web gallery
image_studio_gallery

# Check history with cost
image_studio_history(limit=5)

Style Presets

Preset	Description	Steps	Model
cinematic	Movie-grade lighting, shallow DOF, film grain	30	FLUX Pro
photorealistic	True-to-life natural look, sharp focus	28	FLUX Pro
vintage	Warm faded tones, film grain, kodachrome palette	30	FLUX Pro
fantasy	Epic, magical lighting, rich colors, concept art	32	FLUX Pro
minimalist	Clean, soft lighting, muted palette, editorial	26	FLUX Pro
illustration	Bold artistic style, painterly textures	28	FLUX Pro
noir	High contrast, dramatic shadows, black and white	30	FLUX Pro
studio	Controlled lighting, clean background, product-ready	28	FLUX Pro
gpt-photo	Text rendering, complex scenes, specific constraints	8	GPT Image 1.5
gpt-art	Artistic styles, creative interpretation, stylized	8	GPT Image 2
banana-photo	Photorealism, text rendering, natural-language edits	auto	Nano Banana 2

Social Aspect Ratios

Use these aliases for platform-ready output sizes (FLUX models render exact dimensions):

Alias	Size	For
twitter-post (or x-post, social)	1200x675	Standard tweet image
twitter-header (or x-header, banner)	1500x500	Profile banner
instagram-post (or ig-post)	1080x1080	Feed square
instagram-story (or story)	1080x1920	Story / Reels cover
youtube-thumb (or yt)	1280x720	Video thumbnail

Non-FLUX models (GPT Image, Nano Banana) fall back to the closest standard ratio.

Output Location

Images are saved to /Volumes/Spare Drive/Personal Stuff/Image Studio/ by default. You can change this by setting the HERMES_IMAGE_STUDIO_OUTPUT environment variable in ~/.hermes/.env:

HERMES_IMAGE_STUDIO_OUTPUT=/path/to/your/images

Filename pattern:

YYYYMMDD_HHMMSS_preset_seed_subject.png

Skills

The repo includes 3 Hermes skills in the skills/ directory:

• generate-image.skill.md — Guided single-image generation
• batch-generate.skill.md — Batch variant exploration
• image-studio-workflow.skill.md — End-to-end pipeline from prompt to publish

Copy these to ~/.hermes/skills/ to load them into Hermes sessions.

Design

Zero dependencies. The plugin uses only Python's built-in libraries — urllib, sqlite3, json, os, re. No pip installs, no requirements.txt, no dependency hell. Clone it and it works.

Safe by design. If FAL_KEY is not set, the tools simply don't register. No crashes, no stack traces, no error spam. Set the key when you're ready.

Standard Hermes architecture. Follows the same plugin pattern as the official Spotify and disk-cleanup plugins. Uses plugin.yaml + register(ctx) to wire 5 tools into the image-studio toolset. Predictable, maintainable, and compatible with every Hermes version.

Generation Pipeline

Each image goes through a clean pipeline:

1. Prompt enhanced with the chosen preset's style framing and model selection
2. Sent to the appropriate model endpoint (FLUX, GPT Image, or GPT Image edit via FAL)
3. Downloaded and saved with a descriptive filename
4. Recorded in a local SQLite history database
5. Ready for editing, upscaling, or re-generation with a different seed

License

MIT