Quickstart guide & FAQ
Seven steps from blank page to model-ready prompt. Takes about two minutes once you've done it once.
The dropdown at the top of panel 1 sets the wording style of your prompt — ChatGPT, Meta AI, Nano Banana, Gemini, Midjourney (experimental), or Generic. It never changes your creative choices, only how the prompt talks to the model.
Choose an aspect ratio (1:1, 16:9, 9:16, or Custom to leave it open), tap an artistic style pill (Photorealistic, Cinematic, …), and leave High Resolution on unless you want a lighter touch.
Flip on Use reference images if you'll attach images in your image AI — the tool never sees or uploads them; it just adds wording telling the AI to use your attachments. Add Strict Feature Lock to ask the AI to preserve recognizable identity and features.
New here? Hit ✦ Surprise Me first for a fresh collision — see what the tool can do, then tweak any choice. Or pick one of 8 collision types (Sacred Ritual, Day Job, Mistaken Identity, Shared Dream, The Heist, Wrong Habitat, The Meeting, The Machine) — this is the framework that explains why your concepts belong together. Then fill the three concept slots: an Object (toaster, lighthouse, …), a Creature (octopus, dragon, …), and a World (Victorian London, deep-sea trench, …). Your custom text, target AI, and toggles are never overwritten.
An untouched slot or ingredient adds nothing to the prompt. If you want the image AI to decide something, choose Random — let the AI decide explicitly. Blank and Random are different choices.
Open the ingredient groups and add a reality twist, a mood, or a signature odd detail — each one bends the collision without breaking its logic. Contradictions are welcome; the engine renders them sincerely.
Copy Prompt grabs the model-ready text; Copy Title + Prompt adds the creative title on top. Paste into your image AI (and attach any reference images there). The title is yours for browsing and sharing — it's never printed on the image.
Two or three concepts that have no business being together — a toaster, an octopus, ancient Egypt — forced into one fully coherent image. The tool doesn't just list your picks; it invents the reason they coexist (a religion, a day job, a heist, a dream) and writes the prompt around that story. The stranger the combination, the more sincerely it's rendered.
Each slot draws from its own bank: Object (24 everyday things), Creature (24 beings), World (18 places and eras). Pick one per slot, choose Random — let the AI decide to delegate a slot to the image AI, or leave a slot blank to skip it. The collision framework then weaves whatever you picked into a single scene.
They're the story logic. Sacred Ritual makes your concepts a religion; Day Job makes them coworkers; Mistaken Identity puts one in disguise as another; Shared Dream uses dream logic; The Heist stages a theft; Wrong Habitat shows something thriving where it shouldn't; The Meeting convenes a bureaucracy; The Machine builds them into a contraption. Same three concepts, eight completely different images.
The prompt always carries your ratio choice — 16:9 is the default, and switching to Square, Portrait, or Custom updates the wording (Custom leaves the ratio open and lets the image AI decide). But the ratio is a request in the prompt text, not a command: your image AI's own size setting has the final word, and some models ignore prompt-text ratios entirely. For a guaranteed ratio, also set the size/aspect control inside your image AI to match.
No. It's a prompt generator — it writes copyable plain text you paste into your own image AI (ChatGPT, Midjourney, Gemini, whatever you use). No images are uploaded, processed, or generated here.
No. The reference toggle only inserts wording instructing the receiving AI to use images you attach over there. Nothing leaves your browser except the prompt text itself.
Identity and defining features — recognizable faces, markings, and characteristics from your references or descriptions. It does not freeze the pose, setting, expression, or camera angle; those stay free unless you specifically ask for them.
Blank means the ingredient is skipped entirely. Random is a deliberate instruction telling the image AI to choose that ingredient itself. Use blank to keep prompts short, Random to invite serendipity.
Never. It re-rolls the collision, the three concept slots, and the ingredient checkboxes, but your exact lettering, additional directions, target AI, aspect ratio, style, and toggles are preserved.
Browsing, organizing, and sharing. It's generated from your selections but is not an instruction to put that text on the image. Want words on the image? Use the exact lettering field.
Midjourney interprets structured prompts, reference details, lettering, and aspect-ratio directions differently from the chat-style image AIs. The wording is a starting point — expect to adjust parameters inside Midjourney.
The tool's fine — your image AI is repeating itself. When you generate several images in the same chat, the model starts copying its own earlier work. Open a fresh chat window, or try the prompt in a different image AI, and you'll get fresh results.
Your image AI stopped reading. After several images in the same chat, the model can latch onto a picture it made earlier and keep re-rendering that scene — even when your new prompt describes something completely different. Open a fresh chat, paste the same prompt, and you'll get the image you actually asked for.
Yes, but expect more scrolling. Best experienced on desktop or tablet.