Generate a Featured Image

What it does: Draws a unique image matching your article, uploads it to the Media Library with proper title/alt text, converts it to lightweight WebP, and sets it as the featured image — one click.

Applies to Antradus AI Lite, Publisher & Studio v2.7.0 · Last updated 2026-08-26

Before you start

An image-capable provider with an Image Model selected: OpenAI (gpt-image / DALL·E), Google Gemini (Imagen), or OpenRouter (Flux and compatible models). Anthropic and DeepSeek don’t offer image APIs — you can write with them and switch provider just for images.

Step by step

1. Write or generate your article first — the AI designs the image from the article.
2. In the panel’s image section, optionally add art direction in the instructions box: "Warm Mediterranean tones, cinematic lighting, shot from a low angle".
3. Pick a style preset (see below) if you want a consistent look.
4. Click generate. The image appears in the panel.
5. One click sets it as the featured image. It’s already in your Media Library with a descriptive title, caption, and alt text (good for SEO and accessibility).

Style presets

Ten presets tune the visual language to your niche: Default, Gaming, Medical, News, Sports, Finance, Tech, Food, Travel, Entertainment.

Every preset’s prompt text is fully editable under Antradus AI → 🖼️ Image Generation — rewrite them to match your brand’s visual identity once, and every image follows.

Why every preset describes a photograph

Every preset asks for a photograph, never a banner, thumbnail, poster or "layout" — and that wording matters more than it sounds.

Ask an image model to design a banner, and it does exactly that: it composes a wide picture, then fills whatever canvas is left over with a flat block of colour. You get a photo floating between two cream bars. Ask for "decorative shapes" or "clean empty space on one side" and it lays a swoosh or a curve across the picture, or leaves a third of the frame blank for a headline that never arrives.

So the presets describe scenes, and a hard rule on every image demands the photo fill the entire frame, edge to edge — no borders, bars, frames, panels, rounded corners, or decorative graphics laid over or beside it.

If you edit a preset, keep it that way. Words like banner, thumbnail, poster, layout, template, decorative accent or empty space for text will fight the rule and can bring the bars back. Describe the picture you want to see, not the graphic you’d design around it.

Why there’s never text on the image

Image models can’t spell. Ask one for a headline and you get letter-shaped scribble burned into the picture, which makes a real site look broken — and you can’t remove it afterwards.

So generated images contain no text at all: no headlines, captions, badges, tickers, scores, ratings, logos or watermarks. This is enforced at generation time, so it holds even if you edit a preset. Put your headline on top of the image in your theme or page builder instead, where it stays sharp, selectable and translatable.

Optional color cast

An optional color overlay (settings → Image Generation) tints all generated images toward your brand color — subtle way to make a whole blog feel consistent.

Image shape

Under Antradus AI → 🖼️ Image Generation → Image Shape:

Square is the default only because it’s the shape nothing can get wrong. If your theme shows featured images as a wide banner — most do — switch to Landscape once and every future image fits.

⚠️ Honoured exactly on OpenAI and Google Gemini. On OpenRouter the shape is asked for in words, so some models follow it and some ignore it. DALL·E 2 only draws squares and stays square whatever you pick.

On OpenAI, a non-square image costs slightly more than a square one — it’s simply a bigger picture.

Image quality & what it costs you

Under Antradus AI → 🖼️ Image Generation → Image Quality & Cost:

Each step up multiplies what your provider charges per image, so if you generate images in bulk this is the single biggest lever on your bill. Draft is the default and is more than enough for a blog featured image; move up when you’re making a hero image someone will look at closely.

Sites installed before v2.3.0 started on Let the AI decide and keep that choice on upgrade — nothing is changed behind your back. If you never picked deliberately, switching to Draft cuts both your bill and the wait, and is the first thing to try against a 504.

⚠️ This applies to OpenAI only — it’s the one provider with a quality dial (gpt-image-* uses Draft/Standard/Best; DALL·E 3 maps to its own standard/HD; DALL·E 2 has no quality setting). The setting is greyed out unless OpenAI is your active provider. On Google Gemini (Imagen) and OpenRouter every image costs the same whatever you pick here — there you save money by choosing a cheaper image model instead.

If images never appear and you see a "504 Gateway Time-out"

The plugin explains this one for you where it happens: instead of raw server code you get a plain message and a ⚙ Open Image Quality settings link that takes you straight to the fix.

That error comes from your web server, not from the AI: the image took longer to draw than your host allows a page request to run, so the connection was cut before the plugin could hand the image back. The image itself is usually generated anyway — check your Media Library before you regenerate and pay twice.

Fixes, cheapest first:

1. Set Image Quality to Draft or Standard — the most common cure, and it costs you less too.
2. Pick a faster image model (newer, larger models take noticeably longer). On Gemini and OpenRouter this is the only lever, since they have no quality setting.
3. Ask your host to raise the PHP request timeout (fastcgi_read_timeout / proxy_read_timeout to 300s). If you’re behind Cloudflare, note it caps requests at 100 seconds on Free/Pro plans regardless.

The same applies to albums: nine images means nine chances to hit the limit, so Draft or Standard is what makes a big album finish.

Image format & size

Under Antradus AI → 🖼️ Image Generation → Image Format:

WebP conversion needs the PHP GD extension (almost every host has it). If images save as PNG unexpectedly, ask your host about GD.

Tips & gotchas

• Regenerate as often as you like — each result is a new Media Library item; delete the rejects.
• The image prompt is designed from your article content, so generate the article first for on-topic images.
• Need multiple images through the article? That’s Album Generation.