Img Prompt

Img Prompt's whole thesis is that most people don't need another model — they need a better starting blank page. The three-layer architecture is the actual product: Subject narrows the domain (character, environment, light and shadow effects, art style), Category narrows the facet (eye expression, hairstyle, body type, wing), and Tags supply the vocabulary. That mapping is what makes it faster than typing, and it's the part a weekend clone would get wrong — assembling 5000+ bilingual tags across 18 subject groups and 30+ categories is bulk curation work, not an engineering trick. Two details punch above their weight. First, 'Translate & insert' converts a tag you choose in your own language into English inline, so a Spanish or Japanese speaker ends up with an English-ready prompt without leaving the tool. Second, the prompt box's 0 / 380 character counter keeps you inside a tag-built prompt's natural length — a constraint that, in practice, discourages the rambling prose that models handle worst. The honest weaknesses all follow from the same focus. The tag library is static: nothing learns from your previous prompts, and there is no suggestion engine. The 380-character ceiling is a hard stop if you write long free-form prompts. Midjourney's --ar, --stylize and --no flags aren't surfaced in the editor, so you type them yourself at the end. And there's no visual feedback loop — no gallery of community results the way PromptHero or Lexica offer — so if you're stuck on ideas rather than on wording, this tool won't unstick you. Judge it against its real competition: it isn't competing with image generators, it's competing with a blank text box and a memorized syntax cheat sheet. Against those, tag-picking wins on speed for beginners and non-native English speakers, and ties for everyone else. The self-hosting path via GitHub deployment docs is what makes it a keeper for privacy-conscious users — prompts never leave the machine, and there's no account to manage.