Write it there, paste it here
Nothing to install. Ask ChatGPT or Claude for an English prompt, then paste it into the prompt box on the generate page.
A request you can copy as-is:
Write an English prompt for an image generator. Output the prompt only — no explanation, no quotes. Subject: a woman walking through a Japanese shopping street at dusk Include: - subject detail (clothing, expression, posture) - place and time of day - light direction and quality - camera angle and lens - only things that can be drawn (no smells, no sounds) Comma-separated concrete terms, under 120 words.
The last two lines matter. Diffusion models follow comma-separated concrete terms far more reliably than flowing prose, and words like "the aroma of street food" are simply unrenderable — they take up room and produce nothing.
When you want Japanese signage
EGAKU detects Japanese scenes and repaints the signage with its own model. Even so, naming the text explicitly is more reliable. Left unspecified, signs can come out as Japanese-looking shapes that are not real characters.
Specify 1-3 signs with the exact Japanese text. e.g. a sign that reads 「たこ焼き 六個 三百円」
In testing, a large sign facing the camera comes out with zero errors. Small angled signs degrade, so pick the one to three that carry the shot and let the rest blur — which is what a real photograph does anyway.
Ask for three directions, not one
Ask an AI for one prompt and it tends to play safe. Ask for three different directions, make all three in EGAKU, and compare.
Write three prompts for the same subject with different moods: 1. natural, like a photograph 2. dramatic, like a film still 3. soft illustration
When one lands, give that prompt back to the AI and ask for a small change in that direction. Two or three rounds get you closer than writing from scratch.
ChatGPT and Claude have their own usage policies, and EGAKU has its content rules — check them before working with real people, for example.
Summary
- Prompts → write them in ChatGPT / Claude, paste them here
- Not sure → ask for three directions, make them, compare
- Japanese signage → name 1-3 signs explicitly
- How to ask → comma-separated concrete terms, drawable only, under 120 words
Try one. → Make one free