# 5 · Review: look at every image where it lives You tend to judge an image by its prompt, its file name or a thumbnail. A designer judges it on the page, at real size, next to everything around it, on a phone and in dark mode. Do the same, before you say done. --- ## Step 1 · Look at it in place **Rule:** Render the page and look at screenshots. Open each generated image at 100% as well (`2-generate.md` P5). Look at: - widths **375** (phone) and **1280** (desktop); - **light and dark** mode, if supported; - the share image at about **400px** wide, and the favicon at **16px**. If you cannot render or view images, say so clearly in your report. Never claim an image looks right without having seen it. ## Step 2 · Run the check Run `checks/images.js` on the rendered page at both widths. Fix each finding, or say why it's intentional. File sizes only show when the browser exposes them; the check says when it can't tell. ## Step 3 · Five lenses Go through each lens. Fix what you find **before** reporting. ### R1 · The visitor: does each image earn its place? - [ ] Each image has one job you can name (`1-brief.md` B1). Covering it loses something. - [ ] The image is about this product, place or person, not a generic scene. - [ ] Text over an image sits on a calm area and is readable at both widths. - [ ] Nothing important is only inside an image (`2-generate.md` P4). > **In ihumane Pro** (https://ihumane.ai/#pricing): > - R2 · The designer: did one hand make all of it? > - R3 · The sceptic: is it true, and may we use it? > - R4 · The person who can't see it well > - R5 · The person on a slow phone ## Step 4 · Report honestly Start with the outcome, then: 1. **What you added or changed**, in one or two sentences. 2. **Where each image came from:** real, made, stock (with licence) or generated (with the recipe saved where). 3. **Assumed:** guesses from your brief, such as the style or a placeholder waiting for a real photo. 4. **Not done or not verified:** for example "couldn't see the share card in LinkedIn's preview" or "file sizes unknown for images from the CDN". **Agent trap:** reporting "added a hero image and icons" when two icons came from another set and the real photo is still a placeholder. Lead with what the human must do or decide.