Lesson Content

Evaluating a generation result.

After generation, evaluate the clip specifically against the intended visual target – not just whether it looks good generally, but whether it matches what you were trying to achieve. Identify specifically what is different: wrong subject, wrong lighting, wrong motion, wrong style, wrong mood? This specific evaluation drives the next prompt adjustment.

Changing one variable at a time.

When iterating, changing one aspect of the prompt at a time (adding a specific lighting description, changing the motion descriptor, removing a quality modifier) allows clear attribution of what effect each change had. Changing many elements simultaneously makes it difficult to understand which change produced which difference in the output.

Prompt adjustment vs. fresh approach.

After 3-4 iterations without meaningful progress toward the target, consider whether a fundamentally different prompt approach (not just refined version of the same prompt) might produce better results. Sometimes the initial prompt framing is taking the model in a direction that iterative refinement can't fully correct – a fresh approach from a different angle may unlock better results faster.

Saving effective prompts.

When a prompt produces results well-aligned with the intended outcome, save the full prompt text as a reference. This allows reproducing similar results later and provides a starting point for similar projects, rather than reconstructing effective prompts from scratch each time.