Reve 2.1

Reve 2.1 is an image remixing model that transforms your existing photos into polished new creations using simple text instructions. Instead of starting from a blank canvas, you feed it one or more reference images and describe the changes you want—Reve 2.1 then blends, edits, and recomposes those inputs while faithfully following your directions. It is built for creators who need precise control over layout, composition, and on-image text, making it a natural fit for designers, marketers, product photographers, art directors, and content creators who want to iterate quickly on visual concepts.

At its core, Reve 2.1 excels at three things: strong prompt adherence, layout intelligence, and accurate text rendering. Prompt adherence means the model actually does what you ask, honoring detailed multi-step instructions rather than loosely interpreting them. Layout intelligence means it understands how elements should be arranged within a scene—where objects belong, how they relate spatially, and how to place new items believably into an existing composition. Accurate text rendering means that when you ask it to add words, labels, or names to an image, they come out legible and correctly spelled, which is notoriously difficult for image models and a standout strength here.

A distinctive feature of Reve 2.1 is its ability to work with references and spatial guidance. You can provide between one and eight reference images and point the model to specific ones directly from your prompt. This lets you combine elements from multiple sources—for example, taking a product from one image and placing it into a scene from another. You can also guide the model using visual cues within your input, such as boxes that mark exactly where an object should go or where text should appear. In practice, this means you could instruct the model to drop a specific product into a marked area of a showroom scene and add a brand name inside a designated white box, and it will place both accurately according to your guidance.

The model produces high-quality images and supports a wide range of aspect ratios so your output fits whatever medium you're working in. Options span from ultra-wide cinematic and banner formats like 4:1, 3:1, 21:9, and 2:1, through standard landscape ratios such as 16:9, 3:2, and 4:3, to square 1:1, and tall portrait and vertical formats including 4:5, 3:4, 2:3, 9:16, and even 1:2, 1:3, and 1:4 for extreme vertical layouts. If you're not sure which to choose, an automatic setting lets the model pick an appropriate aspect ratio based on your request. This flexibility makes it easy to produce social media posts, product listings, marketing banners, poster mockups, and more without cropping or reworking your composition.

Reve 2.1 can generate up to four images at once from a single prompt, giving you multiple variations to compare and choose from. You can save your results in common formats including PNG, JPEG, and WebP, so you can pick the option that best suits transparency needs, file size, or downstream editing. On the input side, the model accepts a broad set of image formats—PNG, JPEG, WebP, AVIF, and HEIF—so most photos straight from your camera, phone, or design tools will work without conversion.

Who benefits most? Product and e-commerce teams can composite items into staged environments and add clean, correctly rendered labels or brand names. Marketers and social media managers can rapidly remix existing assets into on-brand visuals across multiple aspect ratios. Designers and art directors can prototype layouts and product placements with precise spatial control. Photographers and retouchers can extend or transform scenes—swapping backgrounds, adding objects, or dressing subjects—while keeping the overall composition believable. Because the model reliably renders text, it's especially valuable for anyone who needs typography baked directly into an image, such as packaging concepts, signage, posters, and promotional graphics.

When writing prompts, be specific and descriptive—Reve 2.1 rewards clear, detailed instructions. Because it follows directions closely, you can layer multiple requests into a single prompt, such as placing an object, adjusting its position, and adding text all at once. Using visual guides like boxes helps the model understand exactly where you want elements to land, which improves placement accuracy for complex compositions. When combining multiple references, clearly indicate which image each instruction refers to so the model pulls the right elements from the right sources.

A few practical considerations: each reference image should be under 10 MB, and the model works within a maximum input resolution, so extremely large source files may need to be resized. You can supply as few as one and as many as eight reference images, giving you room to combine several sources for richer composites. Prompts can be quite detailed, offering plenty of space to describe complex, multi-step edits. Since the model is designed around remixing rather than pure text-to-image generation, at least one reference image is always part of the workflow—this is what makes it so effective at preserving and transforming real visual material.

In short, Reve 2.1 is a precision tool for creators who want dependable, instruction-driven image editing. Whether you're compositing products, building marketing visuals, prototyping layouts, or adding clean typography to a scene, its combination of strong prompt following, spatial awareness, and accurate text rendering helps you turn existing images into finished, professional results.