A generated image is a starting point that looks like a finished one. It reads well on a phone, and then it goes into a layout at 300dpi and the texture falls apart, the hand has four knuckles, the logotype is nonsense and the perspective disagrees with the rest of the campaign. This course is about the distance between that first result and a file that survives a client review, a printer and a legal check.
Day one is direction. You start with enough of how image models behave in practice to work deliberately: what a prompt, a seed and a reference image each control, and why two runs of the same prompt differ. You compare Adobe Firefly, the generative features inside Photoshop, Midjourney and open models that run on your own machine, judging each on what it is actually good at rather than on which is fashionable. Licensing is treated as a design constraint, not a footnote: you go through what each vendor states about its training data and its indemnification, and you learn what to check in your own subscription and your client contract before generative work goes near paid advertising. The afternoon is art direction through language: describing subject, composition, lens, lighting and material rather than naming a style, controlling what stays out of the frame, using reference images for style and structure, and reading your own results critically enough to tell a fixable flaw from a dead end.
Day two is production, which is where most of the value is. You bring generations into Photoshop as an editable layered file, then repair what the model got wrong: hands, type, perspective, texture and grain, using ordinary retouching skills that still decide whether the file passes. You look at what upscaling really recovers and what it only smooths, vectorise a generated mark in Illustrator by redrawing rather than tracing, and bring a generated image into a brand palette and into CMYK without the colour shifting under your hands.
The final afternoon builds a system rather than a picture. You generate variants that hold together across a campaign, decide where generative work belongs in a studio process and where it quietly costs more time than it saves, and inspect the Content Credentials that travel with a file under the C2PA standard. You work through client disclosure, likeness and model releases, and the briefs a designer should decline. The two days close with a campaign set produced end to end and handed over with a written provenance note.
The class runs live online or in the classroom in Colombo, priced in LKR for Sri Lanka and in USD internationally. You keep the exercise files, the prompt log built across both days and the class recording for twelve months.