Reve v2.1: The Essentials

Last updated: July 20, 2026

Reve v2.1 is a reasoning-driven model for precise, controllable image editing. Attach one or more images, describe the change in plain language, and it edits exactly what you asked while keeping every other subject consistent. It also generates from a prompt alone.

Ask it to swap the text on a poster, recolor a product, change a material, relight or restyle a scene, or merge several references into a single image. Because it reasons about the request, it changes only what you name and leaves the rest untouched.

What It Does

  • Precise edits from a plain prompt: recolor, swap text, change materials, relight, or restyle without touching the rest of the image.

  • Text rendering and replacement: reads and rewrites text inside an image, useful for posters, UI, signage, and product labels.

  • Multi-image merge: combine several reference images into one scene while keeping each subject consistent.

  • Consistency: faces, poses, and layouts stay stable across an edit.

  • Text-to-image too: generate from a prompt alone, up to 4000 characters, across many aspect ratios.

Parameters

Prompt

Describe the image or the change you want, up to 4000 characters. To point at a reference image, use a frame tag written as <frame>N</frame>, numbered from 0 in the order you added the references. So the first reference is <frame>0</frame>, the second is <frame>1</frame>, and so on.

Reference Images

Optional images that guide the result. Add one for a single edit, or several to merge subjects or transfer a look. Refer to each one in the prompt with its frame tag.

Aspect Ratio

The shape of the output. Choose a preset from 4:1 down to 1:4, or leave it on Auto to let the model pick a fitting shape. For an edit, Auto keeps the source shape.

How to Use Reve v2.1

Open the model page. To edit, attach the image you want to change, then describe the change in plain language and name what to keep. A reliable pattern is: state what stays the same, then state the single change.

Using <frame>0</frame>, keep the same subject and composition, but [the one change you want].

To merge, add two or more references and describe how they combine, pointing at each with its frame tag. To generate from scratch, add no reference and describe the image you want.

Examples

Each example below is a real generation on Scenario, with the exact prompt used. The first shows a multi-image merge; the rest are single-image before and after edits. Notice how each result changes only what the prompt names and keeps everything else intact.

Merge several images into one

Edit prompt: “Combine the five people from <frame>1</frame> through <frame>5</frame> into one coherent group photo, using a separate layout reference only for positioning. Keep each person's face, hair, and clothing exactly as shown in their own image, everyone outside a cafe with string lights and warm afternoon light, with cohesive shared lighting across all five.”

Five separate portraits, each a different person (inputs)

Merged into one consistent group photo (result)

Recolor and change on-image text

Edit prompt: “Keep the same woman and pose, but change her outfit from red to emerald green and the red LED lighting and background sign to cool teal, with the sign now reading “CALM”.”

Before

After

Restyle into a new medium

Edit prompt: “Keep the same building and composition, but restyle it from synthetic cubism into a glowing Art Nouveau stained-glass window with bold lead lines, jewel-toned glass, and a backlit luminous look.”

Before

After

Transform the environment

Edit prompt: “Keep the same aerial composition and lagoon shape, but transform the tropical scene into a frozen arctic lagoon: turquoise water becomes icy blue, the shore becomes snow and glaciers, floating icebergs, cold pale light.”

Before

After

Swap the subject, keep the style

Edit prompt: “Keep the same shattered pop-art style and composition, but change the animal from a tiger into a howling wolf and shift the vibrant palette to cool icy blues, cyans, and silver.”

Before

After

Tips for Better Results

  1. Say what to keep, then what to change. Naming the constant (“keep the same face and pose”) before the change keeps edits surgical.

  2. Make one change per prompt. Chaining several unrelated edits in one prompt is less reliable than editing, then editing again.

  3. Use frame tags for references. <frame>0</frame> points at your first reference. This is how you tell Reve which image to act on, and how you merge several.

  4. Quote text you want rendered. Put the exact words in quotes so the model renders them verbatim, for signs, titles, and labels.

  5. Be concrete about color and material. “Emerald green pleated skirt” beats “green”, and “brushed steel” beats “metal”.

  6. Leave Aspect Ratio on Auto for edits. It keeps the source shape so the composition is not cropped.

Known Limitations

  • One clear change at a time. Very complex, multi-part edits in a single prompt can drift. Split them into steps.

  • Reference order matters. Frame tags are positional, so if you reorder attachments the numbers change. Double-check which frame points where.

  • Small text can soften. Long or very small text may need a second pass or a larger aspect ratio to stay crisp.

Use Cases

  • Marketing and branding: swap headlines on a key visual, recolor to a campaign palette, or drop a product into a new scene.

  • Game art: reskin UI and HUDs, restyle concept art, or push a prop through seasonal and material variants.

  • Product visuals: recolor a SKU, change a finish, or relight a shot without a reshoot.

  • Content and social: quick restyles, seasonal variants, and localized text versions of one hero image.

  • Compositing: merge a character, a background, and a prop into one consistent frame.