AI Image Generation Arrives on ARMES: Private, Powerful, and Yours to Direct

You have always been able to think with ARMES. Now you can see with it.

Image generation is live on ARMES — the same private, zero-data-retention inference you rely on for text, applied to images. Describe what you want, and ARMES creates it. Your images and references are processed and immediately forgotten. Never retained, never profiled, never monetized.

But we did not ship a text-to-pixels button. We built something better.

Meet ARMES ImageGen

Most image generation tools hand you a text box and wish you luck. The problem is that the gap between what you type and what makes image models excellent is enormous. "A cozy coffee shop" could mean a thousand things. The lighting, the lens, the medium, the composition, the mood — those are the details that turn a generic result into something a professional would be proud of.

ARMES ImageGen is a dedicated agent whose entire purpose is to close that gap. It is a creative collaborator: part art director, part master prompt engineer. You bring the idea — a sentence, a reference photo, a half-formed feeling — and ImageGen brings the craft.

Here is what that looks like in practice:

You say: "A cozy coffee shop at golden hour"

ImageGen creates a prompt like: "A cozy corner coffee shop at golden hour, warm window light raking across a worn oak counter, a single ceramic cup with rising steam catching the light, shot at 35mm with a shallow depth of field, soft film-grain warmth, calm and inviting."

Then it generates the image and shows you both — the exact prompt it used and the result. You are always in the director's chair. If you want the light warmer, the angle wider, the mug gone, you say so, and ImageGen adjusts and regenerates. No guessing. No black box.

The craft, not the text box

ARMES ImageGen is not limited to one discipline. It handles:

• Photography — art-directed photorealism with real camera language: focal lengths, lighting setups, depth of field, film grain. Product shots, portraits, editorial scenes, architectural photography.
• Illustration and art — watercolor, oil painting, vector, anime, concept art, pencil sketch, surreal, abstract. Name the style and medium; ImageGen writes the prompt that makes it precise.
• Typography and design — posters, logos, infographics, packaging mockups, UI concepts. ImageGen understands typographic hierarchy, layout, and how to get legible text rendered inside an image.
• Image editing — attach a photo and tell ImageGen what to change. Swap a background, recolor an object, restyle the whole image as an oil painting, blend multiple references into one composition. It describes only the change, protecting everything you want to keep.

The key insight is that ImageGen does not just forward your words to a model. It translates your intent into the specific, vivid, structured language that modern image models respond to best — and shows you that translation before generating, so you can steer it.

How to create images on ARMES

There are two paths, depending on how much creative control you want.

Path 1: Ask any Pro or Ultra agent

If you are mid-conversation with ARMES GPT PRO or ARMES GPT ULTRA and want an image, just ask. Say "create an image of..." or "generate a picture of..." and your agent will create one. It works from any mode, any model — image generation is available across Pro and Ultra without switching agents.

When your Pro or Ultra agent generates an image, the platform automatically selects the best image model for the request. This is great for quick, inline generation when you are already deep in a conversation and want a visual without breaking your flow.

Path 2: Work with ARMES ImageGen

For dedicated image work — a brand shoot, a series of illustrations, a poster with precise typography, iterating on a concept until it is exactly right — open a conversation with ARMES ImageGen from the Agent Library.

ImageGen gives you everything the inline experience does, plus:

• A creative collaborator who enriches your ideas — it does not just pass through your words; it expands rough concepts into vivid, well-structured prompts using real art direction and photographic language
• Full prompt transparency — you see the exact prompt before every generation, so you can adjust, approve, or redirect
• Manual image model selection — choose a specific model from the launch pool in the mode selector, or let Auto pick the best one for the request
• Iteration as a dialogue — after every image, ImageGen offers concrete next moves: a different palette, a changed angle, a style shift, a tighter crop. Refinement is fast and frictionless
• Consistency across sessions — ImageGen can draw on your notes, saved brand palettes, and prior images to keep your visuals coherent with your identity

Anyone can type a sentence into an image generator. ImageGen turns it into an image a professional would be proud of.

What happens when you generate

When you or your agent generates an image, here is the experience:

1. A placeholder appears in chat showing the aspect ratio of the incoming image while it generates
2. The image renders in chat once complete — click it to view in an expanded lightbox, or download it directly
3. The model that generated it is labeled underneath — you always know which engine produced the result

Generation typically takes a few seconds for standard images, longer for high-resolution or complex compositions. Every standard frame is supported — square, portrait, landscape, and widescreen — up to 4K when detail demands it.

The launch model pool

Image generation launches with four models, each selected for a distinct strength. You can let ARMES choose automatically (Auto mode), or in ARMES ImageGen, pick manually from the mode selector.

• Gemini 3.1 Flash Lite Image — The speed leader. The fastest and most cost-efficient model in the pool, ideal for rapid iteration, quick drafts, everyday scenes, and simple edits. When you are exploring ideas and want to move fast, this is the engine that keeps pace with your thinking.

• Gemini 3.1 Flash Image — The detail workhorse. Higher fidelity than Flash Lite with support for larger resolutions. The right choice when detail matters — fine textures, polished product shots, larger displays — but the heaviest creative triggers do not apply.

• Gemini 3 Pro Image — The precision specialist. Industry-leading in-image text rendering, the best photorealism for people (faces, hands, skin), complex multi-subject compositions, and precise editing. The model you want when "almost right" is not acceptable: posters with readable type, professional portraits, print-ready hero images, and precise edits to existing photos.

• Seedream 4.5 — The artistic powerhouse. The widest stylistic range in the pool — illustration, anime, watercolor, oil painting, concept art, flat design, surreal, abstract. If you want something that looks designed rather than photographed, or you want to creatively blend multiple reference images into one composition, Seedream is where the magic happens.

Auto mode analyzes each request and routes to the model whose strengths match best. It considers the style you are going for (photorealistic vs. artistic), the complexity of the composition, whether text needs to be legible in the image, and the resolution you need. In most cases, Auto makes the right call — and the model used is always labeled under the result so you can learn the pool over time and switch to manual selection when you develop preferences.

This is a living pool. As stronger models ship and meet our quality and privacy requirements, we will add them. The current lineup is always visible on our models page.

What you can make: real examples

To give you a sense of range, here are the kinds of images ARMES users are creating:

• A polished product shot of a watch on marble with studio lighting and sharp reflections
• An anime-style illustration of a character under cherry blossoms
• A restaurant menu design with readable dish names and prices
• A watercolor painting of a quiet canal scene
• A professional headshot with natural skin texture and careful lighting
• A movie poster with bold typographic hierarchy and a stylized title treatment
• A desktop wallpaper at 4K with fine landscape detail
• An edited version of an existing photo with the background swapped to a beach
• A logo for a coffee brand with clean negative space and precise lettering
• A surreal dreamscape with floating islands in an oil-painting style

The range is broad because the craft is in the prompting, not just the model. ImageGen handles the translation from your idea to the right prompt for the right model — and when you work with it directly, you learn the language too.

Same privacy. Always.

Every image you generate runs through the same zero-data-retention infrastructure as everything else on ARMES. The images you create, the references you upload, and the prompts that produce them are processed and immediately forgotten. No model is trained on your images. No provider retains them. No one at ARMES sees them.

This is especially important for image generation, where the inputs and outputs can be deeply personal — a brand identity, a client concept, a personal creative project. On other platforms, those images become training data. On ARMES, they are yours alone.

The privacy guarantee is identical regardless of plan, agent, model, or resolution. It is the same architecture. It is always the same answer.

Available on Pro and Ultra

Image generation is a Pro and Ultra tier capability. If you are on either plan, it is already live — no setup, no upgrade, no toggle.

• Pro users get image generation in ARMES GPT PRO (Auto model selection), plus full access to ARMES ImageGen with manual model selection
• Ultra users get everything Pro has, with the higher rate limits and usage budgets that come with Ultra

If you are on the Free or Eco plan and want to try image generation, start a 3-day Pro trial — cancel before day 3 and you are never charged.

Coming soon: image generation across every Pro and Ultra agent

Right now, inline image generation is available in ARMES GPT PRO and ARMES GPT ULTRA. Soon, it will roll out to every Pro and Ultra tier agent in the ARMES Agent Library — same Auto model selection, same privacy, same seamless experience. Any agent you work with will be able to generate images when you need a visual, without leaving the conversation.

We will share more as the rollout progresses.

This is the beginning

Image generation is a new axis of what ARMES can do, and it is built on the same principles as everything else: private inference, honest architecture, and real capability that respects your sovereignty.

We will expand the model pool as exceptional models ship. We will deepen ImageGen's creative toolkit. We will bring image generation to every Pro and Ultra agent. And we will keep doing what we have always done — giving you the best AI available, from every lab that earns it, without asking you to hand over your data in exchange.

Describe what you see in your mind. ARMES will show it to you.

Open ARMES ImageGen → · See the model pool → · Start your Pro trial →

Joseph Founder, ARMES