Layout-First Architecture
Each image is planned as positioned elements before rendering begins.
Reve v2 AI Image Generator
Reve AI's second-generation model builds an addressable layout of positioned elements before it renders, so a specified arrangement comes back arranged.
- Reve v2
Open this page on a desktop browser to start creating.
The Model Plans the Picture Before It Paints It
Reve v2 works differently from most image models in a way that shows up directly in the output.
Rather than interpreting a prompt straight into pixels, the Reve line plans each image as a structured layout first — every element carrying a position, a size, and its own local description — and renders from that plan. The consequence is that a brief specifying where things go tends to come back with things in those places, which is why prompt adherence, in-image text, and native high-resolution output are the three capabilities this line is consistently singled out for.
Use Reve v2 when the arrangement is the requirement. Build specified compositions, diagrams, product layouts, and any image where "contains exactly these things in exactly this relationship" matters more than "looks appealing".
What Is Reve v2?
Reve v2 is the second-generation image model from Reve, sometimes referred to as Reve AI or Reve Image. It generates from written descriptions with no tier selector and no quality grade.
The model is built around three major strengths:
- Layout-first generation, planning positioned elements before rendering
- Prompt adherence, with named elements appearing as named
- In-image text rendered as part of the planned layout
The line has placed near the top of public text-to-image rankings, and its defining architectural idea — replacing an interpreted prompt with an addressable structure — is also what makes its output more repeatable than models that rewrite your brief internally before generating.
Reve v2 Specs at a Glance
Positional Adherence
Elements described in a spatial relationship come back in that relationship.
In-Image Text
Type handled as part of the planned layout rather than added at the end.
No Tier Selector
One configuration, nothing to choose before generating.
Repeatable Output
Less internal rewriting means less variance between runs of one prompt.
Text-to-Image
Generation from a written brief; editing is handled elsewhere in Virse.
Key Features of the Reve v2 AI Image Generator
Structure Before Pixels
Planning an addressable layout first is what makes a brief binding rather than suggestive. Elements are placed because they were assigned positions, not because the model happened to arrange them that way.
Spatial Relationships Honoured
"Three objects in a row with the fourth balanced on the leftmost" is a specification this model can execute. Most generators treat that as a mood and return four objects somewhere.
Unusual Combinations Attempted
Requests that other models quietly normalise into something more conventional are more likely to be taken at face value here, because the plan is built from what you wrote.
Repeatability Across Runs
Models that rewrite prompts internally produce a different interpretation each time. Less rewriting means running the same brief twice returns two versions of the same idea.
Text Placed, Not Pasted
Because type is part of the layout plan, on-image text sits in a considered relationship with everything else rather than being dropped onto a finished picture.
Precision Is the Trade
Vague prompts get vague results here rather than attractive inventions. That is the deal: the model will not fill in what you left out, which is exactly why it is useful when nothing should be filled in.
Build with Reve v2 in Virse
Specification-driven work has a review step that other image work does not: someone checks the output against a list. Virse makes that check possible in place. The written specification sits on the canvas beside the render, so verifying that all nine elements are present and correctly positioned happens by looking from one to the other rather than by remembering what was asked for.
Keep the Element List Beside the Render
Park the specification next to the image so review is a comparison, not a recollection.
Track What Was Missing
When something is absent, add the clause and generate again next to the previous attempt, so the fix is visible rather than assumed.
Access 30+ Creative Models
Move between Reve v2 and 30+ other image and video models without leaving the canvas or rewriting the brief.
Send Layouts On for Editing
Take a correctly arranged result into FLUX Kontext when a single element needs changing without disturbing the arrangement.
What Can You Create with Reve v2?
Specified Arrangements
Compositions where object positions and relationships are dictated rather than suggested.
Diagrams and Cutaways
Structural illustrations whose parts have to sit in the right places.
Product Layouts
Multiple items arranged to a defined plan.
Deliberately Unconventional Images
Combinations that other models would smooth into something more expected.
Editorial Concepts
Ideas that depend on a specific visual relationship to read at all.
Repeatable Series
Sets where running the brief again should return the same idea, not a new interpretation.
How to Use Reve v2 in Virse
List Every Element
Write down each thing that must appear. Anything left out is genuinely left out.
Give Each One a Position
State where it sits, relative to the frame and to the other elements.
Check the Output Against the List
Compare item by item rather than judging by overall impression.
Add the Missing Clause
When something is absent, it is usually because it was not specified rather than because it was ignored.
How to Write a Reve v2 Prompt
A useful Reve v2 prompt usually includes three elements:
- Every Element
- Its Position
- Its Treatment
Instead of writing
A desk with some office things on it, nicely arranged.
Write
A pale wooden desk seen from directly above. A closed grey laptop centred. A white ceramic mug to its upper right, empty. Three sharpened pencils in a row along the lower edge, parallel, pointing left. A folded pair of glasses at the upper left corner. Nothing else on the desk. Even overhead light, soft shadows directly beneath each object.
Reve v2 Prompt Examples
Specified Arrangement
Four identical white cubes on a mid-grey seamless background, photographed straight on at cube height. Three cubes in a row along the bottom, evenly spaced with a gap of roughly one cube width between each. The fourth balanced on top of the leftmost cube, aligned flush at the front edge. Single soft light from the upper left. No other objects, no texture, no gradient in the background.
Deliberately Unconventional
A formal portrait of a person seated in an ornate armchair, wearing a business suit and a full deep-sea diving helmet, hands resting on the chair arms. Dark green painted wall behind, single window light from the left, no other props. Photographic, waist-up, centred composition. Entirely straight-faced, with no visual joke or exaggeration in the treatment.
Exact Cutaway
A cross-section diagram of a three-storey terraced house, viewed straight on with the front wall removed. Ground floor: a kitchen on the left, a living room on the right, a staircase between them against the back wall. First floor: two bedrooms of equal width. Second floor: one attic room with a sloped ceiling on both sides. Flat illustration, thin uniform line weight, three flat colours plus white, no shading, no furniture beyond what is named.
Reve v2 vs. GPT Image 2
| Dimension | Reve v2 | GPT Image 2 |
|---|---|---|
| Approach | Plans a layout, then renders it | Follows a long brief directly |
| Strength | Spatial arrangement and repeatability | Multi-constraint instruction following |
| Configuration | None, single setting | Quality grade and output size |
| Behaviour on vague briefs | Returns something vague | Fills gaps from world knowledge |
| Best used for | Dictated compositions and diagrams | Long specifications with many clauses |
| Choose it when | Position is the requirement | Detail volume is the requirement |
Tips for Better Reve v2 Results
Write the Full Inventory
Every object you want has to be named. This model does not decorate a scene for you, which is the point.
Position Relative to Something
"Upper right of the laptop" is placeable. "In the background" leaves the model to decide where the background is.
Say What Is Not There
"Nothing else on the desk" is a real instruction here and it prevents the incidental clutter other models add by default.
Judge Against the List, Not the Look
Review by checking off elements. An image that looks good but is missing the third item has failed the brief.
Reve v2 FAQ
What is Reve v2?
Reve v2 is the second-generation image model from Reve, also referred to as Reve AI or Reve Image. It plans each image as a structured layout of positioned elements before rendering.
What is Reve v2 best at?
Compositions with fixed requirements — specified spatial arrangements, diagrams, in-image text, and unusual combinations — where following the brief matters more than producing a conventionally attractive picture.
Is Reve v2 free, and what does it cost?
Reve v2 is available on Virse's paid plans. Generation draws on a monthly credit allowance rather than a charge per image, and the higher plans make it unmetered within fair use. Current plan rates are listed on the Virse pricing page.
Why does Reve v2 follow prompts more literally?
Because it builds an addressable layout of positioned elements before rendering, rather than interpreting the prompt directly into an image. The plan is constructed from what you wrote.
Can Reve v2 render text in images?
In-image text is one of the line's published strengths. For type-led design work such as posters and wordmarks, Ideogram 4.0 remains the more specialised choice in Virse.
Can Reve v2 edit an existing image?
For prompt-based editing in Virse, FLUX Kontext is the model built for it.
Why did my vague prompt produce a plain image?
That is the expected behaviour. This model does not invent detail you left out, which is what makes it reliable when nothing should be invented.
How do I use Reve v2 in Virse?
Select it from the model list, write out every element you want along with its position, and generate.
Specify It and You Get It
Write the full list, place every element, and let the model build what you described rather than what it assumed you meant.