Image models reward specificity. “A cat” gives you a generic cat; “a ginger tabby on a rain-streaked windowsill, soft overcast light, 35mm, shallow depth of field, melancholic mood” gives you a photograph. This tool takes your rough idea and, using your own OpenAI or Anthropic key, rewrites it into three polished prompt variants tuned for your target generator.
How it works
Choose a provider and model, paste your API key, type your rough prompt, and select the target generator — Midjourney, DALL-E, or Stable Diffusion. The tool sends one direct request from your browser asking the model to enrich your prompt with concrete cues for style, lighting, composition, lens, and mood, and to format the output for the chosen generator (parameter flags for Midjourney, tag lists for Stable Diffusion, natural language for DALL-E). It returns three distinct variants so you can pick a direction.
Your key never reaches a Gera server — it is held only in the tab and sent straight to the provider (with the official direct-browser-access header for Anthropic). Refreshing clears it.
What gets added to each dimension
Style layers in the medium (photography, oil painting, digital illustration, watercolour), the period or art movement (Art Nouveau, brutalist, hyperrealistic), and optionally a reference aesthetic (cinematic, editorial, architectural render). A generic “futuristic city” becomes “a neo-brutalist skyline rendered as a Ridley Scott-era science fiction film still.”
Lighting and mood add the light source direction (side-lit, back-lit, rim-lit), quality (soft diffused, harsh directional, golden-hour warmth), colour temperature (cool blue, warm amber), and overall atmosphere. Lighting is often the single biggest determinant of whether an image feels professional or flat.
Composition and lens specify the camera angle (low angle, overhead, eye-level),
framing convention (rule of thirds, centred, negative space), and for photography
prompts, the equivalent focal length and aperture (35mm at f/1.8 for shallow depth
of field; 85mm portrait; wide 16mm for environmental context). These cues travel
well to Midjourney’s --ar parameter for aspect ratio.
Generator-specific formatting
Each generator has conventions the refiner follows:
- Midjourney: Natural language description followed by parameter flags
(
--ar 16:9 --style raw --v 6). The refiner outputs prompts in this format. - Stable Diffusion: Comma-separated weighted tags (
(dramatic lighting:1.3), bokeh, hyperdetailed, 8k). The refiner formats output as a weighted tag list. - DALL-E: Full natural-language sentences without special syntax, since DALL-E 3 follows instruction-style prompts more reliably than tag lists.
Tips
- Keep your input focused on the subject; let the refiner add the descriptive scaffolding around it.
- Generate three variants, render all three, then pick the best phrasing to combine into a final prompt iteration.
- Cheaper models (gpt-4o-mini, claude-haiku) handle prompt rewriting accurately and keep the API cost per refinement negligible.
- After you find a winning prompt formula, save it as a template and swap out just the subject for future images.