Image Agent

Ad images by conversation: describe it, look at it, then simply say brighter, closer, more premium, and watch the brief compound.

Image Agent turns ad-image creation into a conversation. Rather than filling in a form and hoping, your users describe what they want, look at what comes back, and say "brighter", "closer crop" or "new backdrop" — and every follow-up inherits what was already established, so the brief compounds instead of resetting. Behind the chat sits a real art-direction layer: 12 styles, 8 lighting setups, 9 moods and 8 palettes compile into a structured prompt brief rather than a vague sentence.

Brand lock pins every render to a saved brand's colours and tagline and composites the real logo file on engines that accept image inputs, so output is on-brand by default rather than by luck. Users can request up to four variations per turn, switch image engine mid-thread, and pick from four placement-accurate formats. Renders run inline where the request budget allows and quietly hand off to the scheduled processor when they don't, so the plugin behaves on shared hosting with no queue worker running.

Key features

• Conversational refinement — every message is a turn, follow-ups keep prior context, with up to 100 render turns per conversation and a 40-conversation history rail.
• 12 one-click tweaks — More minimal, Brighter, More premium, New backdrop, Closer crop, New angle, Add props, Bolder colors, Softer mood, Add human touch, Surprise me and Regenerate.
• Art-director controls — 12 styles, 8 lighting setups, 9 moods and 8 palettes compiled into a structured brief.
• Brand lock — colours, tagline and the actual logo image carried into every render.
• Up to 4 variations per turn, each an independent render, so users pick a direction instead of accepting a single guess.
• Model picker front and centre — switch image engines mid-conversation without losing the thread.
• Four format chips — 1:1, 4:5, 9:16 and 1.91:1 — matched to real placement specs.
• Optional LLM prompt refiner that rewrites the brief like a creative director would, with a deterministic fallback when none is configured.
• Shared-hosting safe — long renders defer to the scheduled processor instead of timing out.

Powered by

Image Agent is engine-agnostic and ships with five image vendors: Google Gemini (Gemini 3.1 Flash Image, also known as Nano Banana 2), OpenAI (GPT Image 2), Black Forest Labs (FLUX.2 Pro), Ideogram (Ideogram 3.0) and Recraft (Recraft V3). The optional prompt refiner runs on your choice of OpenAI, Google Gemini, Anthropic Claude or xAI Grok. Keys are entered once by the admin and shared across all users.

Ideal for

• Marketers who think in feedback rounds rather than perfect first prompts.
• Brands that need every asset on-palette and on-logo without manual checking.
• Operators who want to offer several image engines and let users compare them.
• Installs on shared hosting with no persistent queue worker available.