Part 7: Teaching AI That the Next Page Is Still the Same Forest

The more specific Downstream became, the less useful a simple prompt like “extend this forest scene” was. The new opening spread needed the waterfall on one side, the oak tree on the other, the stream connecting them, wildlife placed naturally and the viewer standing on a particular bank.

That meant describing relationships, not just objects. The stream had to enter from the previous scene. The oak tree needed to sit slightly to the reader's right. The viewpoint had to remain consistent. Even a vertical artifact or an awkward seam could break the illusion that two images belonged to the same place.

Specific prompts are really specifications

As the project matured, prompting started to look less like asking for a picture and more like directing a production. Perspective, continuity, composition, lighting, crop and print dimensions all became part of the instruction.

The do-overs mattered too. AI can get 90 percent of an image right and still put the remaining 10 percent exactly where the eye notices it. Fixing seams, extending backgrounds and correcting artifacts are part of the workflow, not evidence that the whole approach failed.

The lesson

The closer a creative project gets to finished, the more precise your AI instructions need to become. Early prompts explore. Late prompts specify.

Try it yourself

When revising an AI image, stop describing the whole scene again. Write a change list: what must stay identical, what must move, what must be added and what defect must disappear. Treat it like instructions to an illustrator.

Downstream is now well beyond the original experiment. The next entries can follow the actual production work as the pages are refined toward a finished book.