Newsletter
Stay Ahead
Occasional, engineering-led notes on applied AI — what we're building, running, and learning in production.
Occasional, engineering-led notes on applied AI — what we're building, running, and learning in production.
Use case: Generate article heroes and social cards at publishing volume
Every published article needs at least a hero image, and most editorial teams need a second cropped version for social cards — yet stock photography is generic, licensed photo libraries get expensive at scale, and a design queue that hand-crafts two images per article becomes the bottleneck on publishing day. Webanto's editorial imagery use case generates both directly from the article's headline and summary, using flux.2-dev — a diffusion model Webanto self-hosts on dedicated GPU hardware under a commercial licence — called through POST /api/v1/images. Because output is billed per image and scales with resolution, a newsroom or content team publishing dozens of pieces a week can generate a hero and a social crop for every article without waiting on a designer or paying per-download stock licensing fees, while keeping final publish approval in human hands.
Every article gets a distinct, headline-relevant hero image instead of a generic or repeated stock photo pulled from the same handful of libraries every other site uses.
No per-download stock licensing fees and no waiting on a design queue — generation can happen the moment a draft is ready for review, not after it clears a separate design backlog.
Social card variants generated alongside the hero in the same workflow, keeping visual style consistent between the article page and its social distribution.
Usage-based cost that tracks publishing volume directly, so a quiet week costs little and a high-output week doesn't require negotiating a new stock subscription tier.
Create a Webanto API key under Settings > API and store it server-side as WEBANTO_API_KEY — image generation should be called from your CMS backend or publishing pipeline, not the browser.
Define two prompt templates keyed off the article's headline and dek: one tuned for a wide hero crop (roughly 16:9), one for a square or vertical social card, both referencing a fixed house style so articles don't look inconsistent day to day.
Hook the generation call into your publishing workflow — for example a step that fires when an article moves from draft to ready-for-review — calling POST https://api.webanto.com/api/v1/images with model set to flux.2-dev and the appropriate prompt and resolution for each crop.
Work out the monthly credit budget: a wide hero at roughly 1.5 megapixels runs about 6 credits, and a smaller social card under 1 megapixel sits close to the 5-credit floor, so 40 articles a month at two images each is roughly 40 x 11 = 440 credits, comfortably inside the Starter plan's 10,000/mo allowance.
Route generated images to an editor review step before publish, checking the image actually matches the article's subject and doesn't contain garbled on-image text, which diffusion models still get wrong fairly often.
Store the approved image and its prompt together in your CMS so a rejected generation can be regenerated with a refined prompt rather than starting from a blank slate.
Track credit spend against publishing cadence over the first month and adjust plan tier or prompt resolution if a high-output week runs close to the monthly allowance.
Only if the prompts are similar. Style consistency — colour palette, illustration versus photographic look — should come from a fixed style clause you reuse across prompts; subject variety should come from feeding in the article's actual headline and key subject each time. Teams that copy-paste one generic prompt across every article get repetitive results — the model isn't the limiting factor, the prompt is.
Diffusion models including flux.2-dev are unreliable at rendering legible text inside an image — expect garbled letters more often than not. Overlay the headline as a text layer on top of the generated image using your CMS or design tool rather than asking the model to render it.
This is a genuine limitation worth flagging: generating a recognisable likeness of a real public figure, or an image that could be mistaken for a real photograph of a breaking news event, carries both a misinformation risk and a legal risk around depicting real people. For hard news and anything naming a real individual, use a real photo, wire-service image, or an abstract, non-likeness illustrative image — not a photorealistic generated stand-in.
It depends on volume and your current subscription tier, but the mechanism is different: stock subscriptions are a fixed monthly fee regardless of usage, while generation is billed per image at roughly 5-6 credits (a credit is $0.01 at overage), so cost tracks your actual publishing output rather than a flat licence fee. High-volume publishers with irregular output tend to see the largest relative saving.
For routine daily content, generated imagery covers a large share of what a stock subscription used to. For flagship investigative pieces, interviews, or anything where an actual photograph of a real subject is the point of the image, you still need real photography or licensed press imagery — generation complements editorial visuals, it doesn't fully replace them.
Shoppers who spot a product in the wild, on someone else's feed, or on a competitor's site rarely know the right keywords to find it in your catalogue. Visual search closes that gap: a shopper uploads
Large media libraries accumulate redundancy fast — the same product shot re-exported at a different crop, a stock photo uploaded twice under different filenames, a banner re-saved at a lower compressi
Reverse image search answers a narrower question than catalogue browsing: given one photo someone just handed you — a screenshot, a competitor's product shot, a customer's snapshot of a damaged item —
Start a 14-day free trial on any paid plan. No credit card required.