Reve 2.1 Remix API by REVE AI

Reve 2.1 — Remix

Developer: Reve AI Model ID: reve-ai/reve-2.1/remix Release Date: July 9, 2026

Overview

Reve 2.1 Remix takes several reference images plus a natural-language prompt and composes them into a single new image at native 4K. You can pull a subject from one picture, a style from another, and a background from a third, then describe how they should come together. Rather than repainting blindly, Reve treats an image as a structured set of regions — an "image as code" approach in which style and content references stay consistent across generations, so you can maintain a look, a character, or a brand system while remixing.

This remix variant is the multi-image entry point to Reve 2.1: you supply between one and six reference images and reference them directly in the prompt. Use it when you want to blend or combine visual inputs — as opposed to generating from scratch (reve-ai/reve-2.1/text-to-image) or modifying a single image (reve-ai/reve-2.1/edit).

The Reve 2.1 Model

Reve 2.1 is an update to Reve's flagship image model that sharpens the model's visual intelligence — how well it reasons about structure, hierarchy, and spatial relationships before rendering. On public leaderboards it ranks second overall while using less than a tenth of the compute of the models ranked around it, and it remains the top-rated 4K image generator.

Its foundation is a "large layout model." Images are represented as structured regions that can be reasoned about much like code, which means controllability is built into the model's visual intelligence rather than bolted on afterward. Practical consequences of this design include:

• Higher visual intelligence — the model plans in finer detail and reasons more accurately about how elements relate to one another.
• Native 4K output — full 16-megapixel images (4096 × 4096 class) with sharp edges and preserved texture, generated directly rather than through an upscaling step.
• Stronger text rendering — improved legibility and extended multilingual support, including foreign scripts.
• Consistent style & content references — the layout foundation keeps aesthetics and subjects coherent across generations, which is what makes reliable multi-image remixing possible.

Key Capabilities

• Multi-image composition — Combine 1–6 reference images into one coherent result, drawing subject, style, palette, or background from different sources as directed by the prompt.
• Prompt-level image references — Point to specific inputs directly in the prompt using XML <img> tags (e.g. <img>0</img> for the first image), so instructions can name exactly which reference each part of the scene comes from.
• Style & content consistency — Maintain a recurring character, product, or brand look across a series of remixes.
• Best-in-class typography — Compose scenes with legible, correctly spelled in-image text across multiple languages and scripts.
• High prompt adherence — Faithfully interprets long, detailed prompts (up to 2,560 characters). Prompts are automatically enhanced by the model.
• Native 4K, no upscaling — Renders true 16-megapixel results directly, holding fine detail at full size.
• Built-in post-processing — Apply background removal and a large library of stylistic effects (textures, light, and color) directly in the same request.

Input

This model takes two creative inputs:

• Images — a list of 1 to 6 reference images, each provided as a URL or a base64-encoded string. These are the sources the model draws from.
• Text prompt — a natural-language description of the desired result (up to 2,560 characters, automatically enhanced by the model). The prompt may include XML <img> tags to refer to specific images by their index in the list — for example, <img>0</img> refers to the first image, <img>1</img> to the second, and so on.

At least one image is required. To generate an image without any reference, use reve-ai/reve-2.1/text-to-image; to modify a single image with an instruction, use reve-ai/reve-2.1/edit. The remaining fields below control aspect ratio, output format, and optional post-processing.

Input Parameters

Parameter	Type	Default	Range / Options	Description
prompt	string	—	≤ 2560 characters	Required. Description of the desired image; may include <img>N</img> tags referencing images by index. Automatically enhanced by the model.
images	array of strings	—	1–6 items; each a URL or base64	Required. Reference images the model composes from. Refer to them in the prompt with <img>0</img>, <img>1</img>, etc.
aspect_ratio	string	auto	auto, 4:1, 3:1, 21:9, 2:1, 17:9, 16:9, 3:2, 4:3, 5:4, 1:1, 4:5, 3:4, 2:3, 9:16, 1:2, 1:3, 1:4	Desired aspect ratio. auto lets the model pick the most appropriate ratio.
resolution	string	4k	4k	Output resolution. Reve 2.1 renders natively at 4K (~16 megapixels).
remove_background	boolean	false	—	Keeps only the central subject and makes the background transparent (post-processing).
effects	array	[]	0–51 items	Optional list of post-processing effects (categories: textures, light, color), applied in order. Each item selects an effect and may override its parameters.
output_format	string	png	png, jpeg, webp	Format of the output image.
enable_base64_output	boolean	false	—	Return the image as a Base64 string instead of a URL (API only).
enable_sync_mode	boolean	false	—	Wait for the result before returning, so the image is delivered directly in the response (API only).

Output

Returns the composed image, either as a URL or — when enable_base64_output is enabled — as a Base64-encoded string. By default the request is asynchronous: the response provides a request ID that can be polled until the image is ready. Enabling sync mode returns the finished image directly.

Output formats: PNG, JPEG, WebP. Resolution: native 4K (~16 megapixels).

Use Cases

• Character & product consistency — Keep the same subject or product across many scenes by supplying it as a reference and remixing new contexts around it.
• Style transfer & blending — Combine the content of one image with the aesthetic of another to produce on-brand variations.
• Mood boards to finals — Merge several inspiration images into a single polished composition.
• Advertising & campaigns — Assemble hero visuals from separate product, background, and style references while holding a consistent look.
• Agentic pipelines — Programmatic multi-reference composition where an agent supplies images and references them by index without a human in the loop.

Pricing

Pricing for this model is driven by what each request actually uses. Every request produces one native-4K image at a fixed base price, regardless of how many reference images you supply; optional post-processing (background removal and stylistic effects) adds to the total. There is no separate resolution or upscaling charge, because 4K is generated natively.

SKUs

SKU	Description	Unit Price
sku_image	Base price per generated image (Remix, native 4K, no post-processing)	$0.24
sku_remove_background_mp	Per megapixel for background removal	$0.0036
sku_effect	Per applied post-processing effect	$0.0048

Formula

cost = sku_image
     + sku_remove_background_mp × 16 × (remove_background ? 1 : 0)
     + sku_effect × number_of_effects

• Base image — every request includes one sku_image charge of $0.24. The number of reference images supplied (1–6) does not change the price.
• Background removal — when remove_background is enabled, it is billed across the full 16 megapixels of the 4K image: 0.0036 × 16 = $0.0576. When disabled, this term is $0.
• Effects — each effect in the effects list adds one sku_effect charge of $0.0048; an empty list adds nothing.

Examples

Scenario	Calculation	Cost
Base remix (no post-processing)	0.24	$0.2400
Background removal	0.24 + 0.0036 × 16	$0.2976
1 effect	0.24 + 0.0048 × 1	$0.2448
3 effects	0.24 + 0.0048 × 3	$0.2544
Background removal + 2 effects	0.24 + 0.0036 × 16 + 0.0048 × 2	$0.3072

A plain remix with no post-processing costs the base $0.24, whether you supply one reference image or six. Background removal and each stylistic effect are the only add-ons, so leaving both untouched yields the base price.