prompt-images

prompt-images is a reference guide for writing effective prompts for AI image generation and editing models on Replicate. Distilled from Replicate's blog posts (2024-2026), it is deliberately model-agnostic, focusing on transferable prompting principles rather than any single model, and it directs users to query the Replicate API for current model choices rather than assuming named models still exist.

The guide covers the craft of prompting in depth: using natural-language sentences over keyword lists, being specific about colors, materials, lighting, and spatial relationships, naming subjects directly instead of using ambiguous pronouns, and writing long structured prompts while accepting that more requirements raise the chance of a missed detail. It teaches photographic vocabulary (film stocks, lens characteristics, depth of field, lighting setups like Rembrandt and rim lighting, and composition rules), techniques for rendering text accurately (quoting desired text, matching character counts, inpainting text regions), style transfer, character consistency across generations, and image-editing principles such as specifying what to keep, choosing precise verbs, object removal, background replacement, and the difficulty of perspective changes.

Target users are developers, designers, and prompt engineers building image-generation features or crafting prompts for Replicate-hosted models. Use cases include improving prompt quality for text-to-image and editing tasks, maintaining a consistent character across a series, and rendering readable text in generated images. It is a pure guidance document with no code execution, credentials, or tooling — it points to sibling skills (find-models, compare-models) for model selection and pricing.