Prompt variation generator
The fastest way to find a great image prompt is to explore the space, not to perfect one line. Given a base prompt, this tool produces ten variations by swapping modifiers along the axes you choose — style, lighting, mood, and camera — so you can generate them as a grid and pick the winners.
How it works
Each axis has a curated library of high-signal modifiers:
variation = base_prompt + style[i] + lighting[i] + mood[i] + camera[i]For every row, the generator selects a distinct combination from the enabled axes, walking through the libraries so the ten results stay diverse rather than repeating. Disable an axis and it stays constant across all rows, which is how you isolate the effect of a single dimension.
What each axis controls
The four axes correspond to the variables that image generation models respond to most consistently and predictably:
Style covers the overall artistic medium and visual language: photorealistic, oil painting, watercolour, concept art, illustration, cinematic, anime. Style modifiers often have the largest single effect on the look of an image, because they shift not just colour and texture but the entire visual grammar the model draws from.
Lighting covers how the scene is lit: golden hour, blue hour, overcast, dramatic rim lighting, neon, candlelight, studio softbox. Lighting changes mood and atmosphere substantially even when the subject and style stay constant. It is one of the most efficient axes for exploring emotional range across a single subject.
Mood covers the emotional tone and energy: serene, tense, melancholic, joyful, ominous, nostalgic. Mood modifiers overlap with lighting and colour, but they direct the model at a higher semantic level — the feeling of the image rather than its technical properties.
Camera covers the implied vantage point and lens: low angle, bird’s eye view, close-up portrait, wide establishing shot, macro, fisheye. Camera perspective is particularly important for character-focused subjects, where the angle changes the power dynamic and emotional register of the image.
Choosing which axes to enable
Enabling all four axes at once produces maximum variety across the ten variations, which is useful for early exploration when you have no strong hypothesis about what will work. Disabling all but one axis is the right choice when you want to understand specifically what that axis does to your subject — for example, running ten lighting variations of the same base prompt to learn how each lights your particular subject.
The most efficient workflow for most users:
- Enable all axes on the first batch to get a broad map of the space
- Identify which direction or two looks most promising
- Disable the axes that are already working well, and vary only the axes you want to refine
Why ten variations and not more
Ten variations represent a practical balance between coverage and cost. At low step counts (20 steps, small resolution), ten images in a grid takes less than a minute in most WebUIs and costs a few cents in API calls. It is enough to establish clear directional preference without so many options that the decision becomes analysis paralysis.
If the ten variations all look similar, the base prompt is too specific — some of the modifier vocabulary is not making a visible difference to your subject. Try broadening the subject or enabling axes you had disabled.
Tips for systematic exploration
- Isolate, then combine. Vary only lighting first to learn its effect on your subject, then turn on more axes once you know what each does.
- Grid then refine. Generate all ten cheaply (low steps, small images), pick two or three directions, and re-run those at full quality.
- Keep the subject stable. Change modifiers, not the noun — that’s how you compare apples to apples.
- Save the winners. When a combination clicks, note the exact modifiers so you can reuse that recipe on future subjects.