01 Core Formula
Grok Imagine 2.0 responds best to editorial, descriptive prompts — full sentences that name the subject, the style and the light. Because the model plans layouts deliberately, telling it what text appears and where it sits pays off more than with most image models.
1.1 The Basic Prompt Formula
Start with three required parts — subject, style and lighting — then layer in what the model is unusually good at: exact text content, layout instructions and mood.
[Subject] + [Style] + [Lighting] + optional [Text] [Layout] [Mood]
Subject
The main person, object or scene. Be specific: 'a raccoon in a teal Hawaiian shirt' beats 'an animal'.
Style
The visual treatment: vintage travel poster, editorial magazine spread, studio product shot, ink wash painting.
Lighting
Golden hour, softbox, overcast window light — light sets the mood fast.
Text
Quote exact lettering in the image: headlines, captions, badges. Grok Imagine 2.0 keeps small text legible.
Layout
Where things sit: 'headline on the left, photo upper right, footer strip at the bottom'. The model plans composition.
Mood
A single emotional word: nostalgic, playful, serene, defiant.
Retro Travel Poster
Prompt
A vintage travel poster for a fictional coastal town called 'Saltmarsh Bay'. Bold stacked headline 'SALTMARSH BAY' at the top in weathered condensed typography, subtitle 'Where the tide writes the schedule' below it. Center: a hand-painted lighthouse on a rocky point at golden hour, warm off-white paper background with subtle grain, teal and sandy beige palette. Small footer strip with three icons: a wave, a crab, a lantern. Nostalgic, calm, collectible.
Subject + style + light first, then exact quoted text and layout instructions — the combination this model handles best. Preloaded at 3:2.
1.2 Text and Dense Layouts
Most image models fall apart on small text and multi-block layouts. Grok Imagine 2.0 is built for exactly that: infographics, itinerary cards, magazine pages and posters with several text zones. Name every text block explicitly and give it a position.
Dense Infographic
Prompt
An educational infographic titled 'HOW COFFEE TRAVELS' with a warm cream paper background. A horizontal journey in four illustrated stages, left to right: '1. CHERRY' (a red coffee cherry branch), '2. DRYING' (raised drying beds under sun), '3. ROASTING' (a vintage drum roaster), '4. BREWING' (a steaming pour-over). Each stage has a one-sentence caption in small clean type. Footer band reads 'FROM FARM TO CUP — ABOUT 4,000 KM'. Muted terracotta and forest green palette, fine line illustration style, every label sharp and legible.
Four text zones, each named and positioned. Preloaded at 16:9 — wide canvases give dense layouts room to breathe.
02 Canvas and Aspect Ratios
Grok Imagine 2.0 keeps the canvas choice simple: five presets — 1:1, 2:3, 3:2, 16:9 and 9:16 — plus an auto mode in image edit that follows your input image.
2.1 Choosing a Ratio
Pick the ratio for the publishing channel: 1:1 for feeds, 9:16 for stories and phone screens, 16:9 for widescreen, 3:2 and 2:3 for print-style posters and covers. When in doubt, let the subject decide — tall subjects read better in portrait, landscapes in landscape.
Story-Format Cover
Prompt
A vertical story-format cover for a mystery podcast: a lone streetlamp glowing in heavy fog on a cobblestone alley, deep indigo night palette. Title 'THE FOG FILES' in large distressed serif type in the upper third, 'SEASON 2' in a small stamped box at the bottom. Cinematic, moody, high contrast.
Preloaded at 9:16 — the story format. Tall composition instructions ('upper third', 'at the bottom') help the model stack the text zones.
2.2 Auto Ratio in Image Edit
In image-to-image mode, auto follows the aspect ratio of your input image — the natural choice when you're editing an existing asset and don't want to reframe it. Pick an explicit preset only when you deliberately want to change the canvas shape.
Edit Without Reframing
Prompt
Keep the exact composition and framing of the uploaded poster, but replace the headline with 'AUTUMN EQUINOX MARKET' in the same hand-lettered style, and warm the color palette toward amber and rust.
Upload your poster in image-to-image mode with the ratio set to auto — the edit preserves the original canvas instead of forcing a new one.
03 Editing with Reference Images
Image edit accepts up to 5 reference images. One image is enough for restyling and text swaps; multiple references help when the output must combine a subject, a style and a scene from different sources.
3.1 Upload and Describe the Change
The simplest edit: upload one image and describe what changes and what stays. Say explicitly what must be preserved — typography style, layout, palette — so the model doesn't redesign the parts you liked.
Poster Text Swap
Prompt
Replace the main title of the uploaded poster with 'WINTER LIGHT FESTIVAL' while preserving the original typography style, layout structure and paper texture. Adapt the smaller subtitle to read 'DECEMBER 12–21 · OLD HARBOR'. Keep every other element — the illustration, badges and footer — unchanged.
Upload your poster in image-to-image mode. Naming what stays unchanged is as important as naming what changes.
3.2 Multi-Reference Consistency
With several references, assign each one a role in the prompt — 'the product from image 1, the lighting from image 2, the background from image 3'. Up to 5 reference images are accepted; more than that starts to dilute rather than help.
Product Meets Scene
Prompt
Place the ceramic mug from the first reference image on the wooden windowsill from the second reference image, lit by the soft morning light in the third. Keep the mug's glaze color and shape exactly as in the reference. Cozy, quiet, editorial lifestyle photography.
Upload the mug, the windowsill and the lighting references in image-to-image mode. Preloaded at 3:2 — switch to auto if you'd rather follow one of the inputs.
04 Limits and Moderation
Grok Imagine 2.0 on Molyin keeps the parameter surface small on purpose: one image per run, five ratios plus auto, no seed, no resolution tiers. Knowing the edges upfront saves you from surprise results.
4.1 One Image per Run, No Seed
Each run returns a single image, and there is no seed parameter — rerunning the same prompt gives you a fresh variation, not a pixel-identical copy. Treat reruns as variations: lock the composition in words, not in a seed number.
Icon Draft Iteration
Prompt
A minimal app icon for a stargazing journal: a small crescent moon above three handwritten lines, deep navy rounded square, gold foil details. Flat vector style, clean geometry, no text.
Preloaded at 1:1. Because every run is a fresh variation, generate a few times and keep the strongest draft — that's the intended workflow.
4.2 Content Moderation
Prompts and reference images pass through safety filters. If a request is blocked, the error code is content_rejected — rephrase rather than retry, because the same input will hit the same filter.
Fantasy Map
Prompt
A hand-drawn fantasy map of a fictional island chain called 'The Ember Isles', compass rose in the lower right corner, sea serpent illustration in the ocean, mountain ranges and forest icons marking three named islands: 'Ashfall', 'Thistledown', 'Kingfisher's Rest'. Aged parchment texture, sepia ink, cartographic illustration style.
Fantasy and stylized content is usually fine. Realistic violence, gore or sexual content will be blocked regardless of artistic framing.