Wildfield generates images inside the ad, onto the layer you are working on, at the size of the format you are in. You describe the picture, click chips to drop in the brand's colours, logo and call to action, and the result lands on the layer, ready for type and motion. Every image you make is kept in the design system's library, so it can be reused in other scenes and formats.
Image generation lives in the Retouch tab. Open it from Canvas with Retouch Image, or the pencil icon on an image layer. This page covers making a new image; Retouch covers changing one.
#Generate
On a layer with no image, the section is called GenAI. Write a Prompt describing the subject, the action and the setting, pick a Model, and press Generate. The new image replaces the empty slot and is added to the library.
Prompt chips
Chips under the prompt insert a placeholder that Wildfield fills in for each scene when you generate:
{c1},{c2}and so on: the brand palette, one chip per colour with its swatch. Each becomes the colour's hex value and name.{logo}: "the (brand name) logo".{cta}: a button reading this scene's call to action, or a generic button if the scene has none.{svg1}and up,{glyph1}and up: the scene's vector graphics and glyphs, when it has any.
Because the chips resolve per scene, one prompt works across a whole campaign. "A shopper holding a tote bag in {c1}, {cta} on a shop window" asks for each scene's own CTA on the window.
Models and sizes
The Model list offers several image models. Their labels say what each is good for (fast, detail and so on), and one makes images with a transparent background, useful for product cut-outs over a gradient. Each click makes one image. Generating spends credits: from A$0.09 to A$0.70 per image, depending on the model.
Images are generated at twice the format's size, so a 9:16 image is sharp at full resolution, and a new image on the base layer fills the frame.
Brand style and composition
If the design system has a Global Image Prompt (set in Input mode), it is added to every new image so the whole campaign shares a look. A Global prompt on switch under the prompt turns it off for one image. Under the prompt, + Character adds a recurring character and + Object a product or prop photo, with Advanced on or off. Advanced asks the model to keep the headline, logo and CTA areas of the layout clear of the subject.
Use this scene's video frame
When the project has a background video, Use this scene's video frame feeds the model the exact frame the scene starts on, framed as you see it in this format. The button becomes Generate from video frame, and the result drops back on top of the footage in register, so you can add an object or change the look of the opening frame. It only uses the video, never your text or other layers.
One format at a time
A generated image belongs to the format it was made on. Generate on 9:16 and the 1:1 and 16:9 versions of the scene keep their own images. That stops a tall image being squeezed into a wide frame. To reuse it, switch to the other format and pick it from the library; its tag tells you which format it was made at, and the pick applies to that format only. To fill empty space around an image in a new format, use Extend Edges on the Canvas layer panel instead.
Generations and Describe
Generations, below Media in the Retouch tab, shows every image made for this design system. Click one to put it on the layer. The arrows in Media step through this layer's earlier images. Describe (the sparkle icon) reads the current image and writes a prompt that would recreate it: a quick way to make variations of a picture you like.
Worked example
A 9:16 sale ad for a sportswear brand, three scenes, one prompt.
- In scene 1, write "a runner mid-stride on a wet city street at dawn, jacket in {c1}, {cta} on a billboard behind".
- Press Generate. The prompt sent asks for a billboard reading this scene's CTA, "Shop the drop".
- Copy the prompt to scenes 2 and 3, whose CTAs are "Last sizes left" and "Ends Sunday", and generate. Each prompt asks for its own scene's words on the billboard.
- Switch to 1:1 and pick the best image from the library, or generate a square version.
Image models do not always spell text perfectly, so check any words in the picture before you export, and regenerate if a letter is off.
Ideas
- Generate a product cut-out with the transparent model and put a particle field behind it.
- Write one prompt with {c1} and {c2} and generate it for every market in a campaign, so colour stays on brand in each one.
- Use the video frame option to add a product to the opening frame of a stock clip.
- Describe a photo from the client's website, then edit the prompt into a fresh scene in the same style.