IMGPrompt: building model-ready English prompts by clicking tags in your language

What IMGPrompt solves

Writing prompts for AI image and video models means knowing the English wording those models respond to, which is a barrier for anyone who does not think in that vocabulary or that language. IMGPrompt removes it. You click tags in your own language, and the app assembles the English prompt that models such as Midjourney, FLUX, GPT-Image and Veo actually expect, ready to copy. The interface is offered in many languages, and the README describes an 18-language site. The audience is creators who use image and video models and want to build prompts quickly from a curated tag library rather than memorizing prompt syntax or translating by hand. It is a prompt builder, not a generator: it produces the text you paste into a model, and the generation happens wherever you take that prompt.

A localized tag picker that outputs English

The mechanism is a tag-and-swatch picker mapped to English prompt fragments. You browse categories of tags shown in your language, select the ones you want, and the app composes them into the English wording the target model understands, so the translation and the prompt syntax are handled for you. Recent releases describe a reworked tag and swatch picking flow, single- and multi-color swatch views, prompt preview thumbnails you can zoom, and desktop niceties such as remembering your last language and following the system language on first launch. The design keeps the human in control of composition, you pick the elements, while the app owns the mapping from a localized label to the phrase a given model expects. That mapping quality is the whole value, since the app itself generates nothing visual.

Self-hosting with Docker

IMGPrompt is a static web app you can use online or run yourself, and the published image needs no build step. To self-host, run the container and map its port:

docker run -d -p 5666:5666 --name imgprompt ghcr.io/rockbenben/img-prompt:latest

Then open the local address on port 5666 and it is yours, which the README notes works on a NAS, a home server or an air-gapped machine. A docker-compose.yml also ships in the repository, and because the app is a plain static export, any static host serves it just as well. There is also a hosted instance you can try online without installing anything. The first real use is opening the app, clicking a few tags in your language, and copying the assembled English prompt into your image or video model.

Where a prompt builder stops

The limitations follow from what IMGPrompt is. It builds prompts; it does not generate images or video, so you still need Midjourney, FLUX, GPT-Image, Veo or another model, and whatever those cost, to turn the prompt into a result. Its usefulness is bounded by its tag library and the mappings behind it: a concept not covered by a tag will not appear, and the English phrasing a model prefers can drift as models change, so the mappings are only as current as their last update. It is also a composition aid, not a guarantee, since the same prompt yields different results across models and runs. None of this detracts from its core job, but it means IMGPrompt is one step in a workflow rather than an end-to-end tool, and it pairs with a generator rather than replacing one.

IMGPrompt versus writing prompts by hand

The alternative is writing prompts yourself, possibly with a prompt-library website for reference. Hand-writing gives you complete freedom of phrasing and no dependency, but it requires knowing the English vocabulary each model rewards and, for non-native speakers, translating intent accurately, which is exactly the friction IMGPrompt targets. A static prompt-library site helps you look up examples but still leaves you to assemble and adapt them. IMGPrompt's difference is that it turns selection into a finished, model-specific English prompt in your own language, and it is self-hostable so it can run privately. If you already write fluent model prompts, you may not need it; if you work across languages or want to build prompts quickly from a curated vocabulary, it removes real friction, with the honest caveat that you still bring the generator.

MIT license and cadence

IMGPrompt is MIT-licensed, so it is freely reusable and self-hostable, and it is a Next.js application shipped both as a hosted site and a Docker image, with the package version in the repository at the 3.x line. The last push was on 2026-09-13, and the release notes describe an active redesign, the reworked picker, swatch views, preview thumbnails and desktop integration, so the tool is being maintained and refined. Because its value rests on the tag library and the prompt mappings, treat freshness as important: use a recent version so the model-specific wording stays aligned with how current models behave. Adopt it as a front-end step in your generation workflow, self-host it if privacy matters, and keep a real image or video model on the other end to actually produce results.

Editorial conclusion

Adopt IMGPrompt if you build prompts for image and video models and want to assemble model-ready English wording by clicking tags in your own language, optionally self-hosted for privacy. Do not expect it to generate anything: it produces prompts, and you still supply Midjourney, FLUX, Veo or another model to render them. Try the hosted site or self-host with the Docker image on port 5666, and keep to a recent version so its model-specific prompt mappings stay current.

Frequently asked questions

Official sources

1. License: MIT
2. Project website
3. README
4. Releases
5. rockbenben/img-prompt on GitHub