ai-product-photographylisted

# AI Product Photography A compact operating doctrine for generating product images that a customer would accept as the real thing. The goal is **Conversion Integrity**: Accuracy (physical truth of the product) plus Realism (passes the sniff test) plus Branding (speaks the brand's visual language). Miss one and the image hurts the brand. This doctrine is diagnosis-first: name what is wrong, pick one technique, commit. When a technique keeps failing, switch methods rather than forcing the same one harder. This is the public, blind-judge-measured version of the system. Where a claim was measured, the numbers are in `references/evals.md`; where a technique needs its full treatment, the link is in `references/techniques.md`. --- ## 1. The foundations (the laws) These describe how current image models behave. Every technique below is an application of one or more of them. - **Small Steps Law.** Never ask the model to make a big jump in one step. A "jump" is the distance between what your inputs already show and what you are asking for. Big jumps fail in random ways. Every technique turns one big jump into small checked steps. - **Words are the strongest input.** Text-to-image is the highest-fidelity path when the thing is describable. The photo is a crutch for what you cannot describe: it helps, but it takes word-level control away. Before attaching a reference, ask "could I just describe this?" - **Say it or show it, never both.** Words for what you want to change, pictures