Mirai · Prompting Guide
Get the most out of Mirai
Read this once. Skim back to it when a generation isn't doing what you expected. Every pattern below is a real lever you control from the chat input.
How Mirai works
Talk to it like a creative director
Mirai understands plain English. Type what you want, reference any image you have, and the system picks the right model angles, lighting, and composition. The patterns below tell you the magic words that unlock specific outcomes — so you stop wasting credits on guesses.
01 — Magic keywords
Words that change the output
Drop any of these into your prompt and Mirai will switch behaviour.
When you upload a finished photo and want your Mirai model swapped in for whoever is currently in it. Everything else — pose, wardrobe, eye masks, jewelry, scene, lighting — stays pixel-identical. Use this for product photography you want to re-shoot with your licensed model.
Reframes the shot without regenerating the scene. Combine with edit follow-ups ("wider crop, show her shoes") to iterate framing on the same image.
Switches which canonical reference angle drives identity. Profile prompts pull the profile shot; 3/4 pulls the 45° shot. Use when face direction matters to the composition.
Sets framing AND auto-bumps resolution to 2K so the face stays sharp at smaller pixel density. Worth the extra ~30s render time on body shots.
Triggers the expressive canonical anchor instead of the neutral headshot. The model's smile/expression in your output matches her canonical expressive shot.
Bumps output resolution. Starter and Growth render at 2K (the default for body shots); Scale + Enterprise unlock 4K. 4K adds ~60s to render time and is meant for print or billboard use — overkill for IG and most digital surfaces.
02 — Reference uploads
When to attach an image (and how)
Mirai handles four upload patterns. Picking the right one in your prompt is the single biggest quality lever.
Product
"Model wearing this silver silk eye mask, lying on white linen pillows"
A flat-lay product photo, a garment on a hanger, jewelry on a card. The product enters the scene exactly as photographed.
Scene / location
"Place her in this hotel suite, late afternoon light through the window"
An empty room, a balcony, a beach. The model is composited into the location with its lighting preserved.
Recreate
"Recreate this image with my model"
A finished campaign photo you want re-shot. Adds the recreate keyword to the prompt; Mirai preserves every pixel except the person's identity. See section 01.
Inspiration / moodboard
"Editorial portrait in this lighting and palette" + click the ✨
A Pinterest screenshot, a Vogue cover, a competitor's campaign. Click the ✨ on the attachment chip to use it as inspiration only. See section 03.
03 — The sparkle ✨ toggle
“Use as inspiration only”
Hover any attached image and a small ✨ chip appears in the corner. Click it to switch that reference from literal to inspiration. Behind the scenes Mirai removes the image entirely and replaces it with a written description of its mood, lighting, palette, and composition before sending to the generator.
Use it when
- You uploaded a Pinterest / Vogue / Instagram image with someone else's face in it (otherwise that face bleeds into your output)
- You want the vibe, palette, lighting, or composition — but NOT the actual person, outfit, or products
- You're building a moodboard reference, not a literal source
Don't use it when
- ·You uploaded a product (eyemask, dress, sunglasses) — you need the exact pixels
- ·You uploaded a location you want preserved
- ·You typed 'recreate' — defeats the purpose
- ·The reference IS the source of truth, not a vibe
04 — Edit vs. fresh start
Every follow-up edits the previous image
In the same conversation, your second message is treated as an edit of the previous generation — “make her smile”, “change the dress to black”, “wider crop” all modify the same image. Mirai preserves face, hair, and everything not mentioned in the edit, so you can iterate without losing identity consistency.
To start fresh with a different scene, open a new conversation from the sidebar. Don't try to pivot mid-conversation — the edit pipeline will fight you.
05 — Aspect ratio
Pick the format before you generate
The size selector under the chat input controls output dimensions. Portrait 4:5 for IG posts, Square 1:1 for catalog, Portrait 9:16 for Stories / Reels, Landscape for hero banners and blog headers.
Recreate mode auto-inherits the source image's aspect — you don't need to match it manually.
06 — Wardrobe vocabulary
Editorial words the safety filter accepts
The image model refuses prompts with explicit lingerie / swimwear / nudity language even in commercial contexts. Use editorial substitutes:
07 — What goes wrong
Common mistakes that waste credits
Uploading a Pinterest face without the ✨ toggle
The reference's face leaks into the output. Fix: click the ✨ on the attachment chip to drop the image and keep only the vibe.
Vague prompts ("a nice photo of her")
The auto-enhancer fills in the gaps but quality is uneven. Add scene, wardrobe, and lighting at minimum.
More than 2-3 reference images
Mirai weights all attachments together; too many dilutes each one. Stick to 1-2 strong references per generation.
Contradictory instructions
"Close-up full body shot" — pick one framing. Same for "editorial casual candid" — pick a register.
Trying to start over mid-conversation
The edit pipeline kicks in on every follow-up. Open a new conversation for a fresh scene.
Stuck on a generation that isn't working? Email support@miraitalent.ai with the prompt and we'll diagnose it.