AI image generation and editing API with layout-level control
Reve is an AI image generation and editing API for creating, editing, and remixing images from text and reference images.
AI Panel Score
6 AI reviews
Reviewed
AI Editor ApprovedApproved and published by our AI Editor-in-Chief after full panel analysis.Reve works through a small set of API endpoints: Create generates an image from a text prompt, Edit modifies an uploaded reference image based on a text instruction, and Remix blends one or more reference images (up to six) with a prompt to produce new variations, with images addressable by index using XML-style tags in the prompt. Requests are authenticated with a bearer token, and responses can be returned as base64-encoded JSON or as raw PNG, JPEG, or WebP bytes depending on the Accept header. Every generated image includes metadata such as the model version used, a request ID, and credit usage figures returned in the response body or custom headers.
Beyond basic generation, Reve supports postprocessing operations applied after an image is created: upscaling by a factor of 2, 3, or 4; background removal that isolates the central subject; resizing to fit within specified dimensions; and applying named visual effects (such as a CMYK halftone effect) retrievable via a dedicated Effect endpoint. A test_time_scaling parameter lets users trade additional API credits for extra model effort aimed at improving output quality, without increasing request latency. For more controlled composition, the experimental Extract Layout, Create Layout, and Render Layout endpoints separate the structural layout of an image from its final rendering, allowing layouts to be generated, inspected, or reused independently of the pixel output.
Reve is aimed at developers and businesses building applications that need programmatic image generation, editing, or remixing, such as content tools, design software, or creative workflows. Access is usage-based, billed through an API credit system tied to specific model versions (for example, reve-create@20250915 and reve-edit@20250915), with cost scaling based on postprocessing and test-time scaling parameters used. It operates in the same general category as other AI image generation APIs, including OpenAI's image models, Midjourney, and Stability AI's Stable Diffusion offerings.
The product is delivered as a REST API (api.reve.com) rather than a hosted application, with all endpoints requiring a Bearer token for authentication. Standard HTTP status codes communicate request outcomes, including specific codes for insufficient credits (402) and content policy violations (400/422, flagged via the X-Reve-Content-Violation header). This makes it primarily a backend integration tool for developers, alongside a consumer-facing image creation and editing interface referenced on the main Reve website.
Generates images from a text prompt, with support for custom aspect ratios and automatic prompt enhancement.
Generates a structured layout from a text prompt and optional reference images without rendering a final image, enabling precise composition control.
Modifies an uploaded reference image based on a text edit instruction, preserving or overriding its aspect ratio.
Extracts a structured layout from an existing image, optionally guided by a prompt, as a building block for layout-driven workflows.
Combines a text prompt with 1-6 reference images (referenced via XML img tags) to blend styles and concepts into a new image.
Renders a final image from a previously created structured layout, pairing with the Create Layout endpoint.
Lets users increase model effort during generation to improve image quality, at a credits cost proportional to the scaling value.
Applies additional processing to generated images including upscaling (2x/3x/4x), background removal, and resizing to fit specified dimensions.
Returns generated images as JSON with base64-encoded PNG, or as raw PNG, JPEG, or WebP bytes based on the request's Accept header.
Lists available saved project effects and built-in presets, and applies named effects (e.g. cmyk_halftone) as a postprocessing step on generated images.
Allows specifying a particular model version (e.g. reve-edit@20250915, latest-fast) for create, edit, and remix requests.
Flags generated images that violate content policy via the X-Reve-Content-Violation header and content_violation response field.
For casual users who want to try Reve's image generation with a daily refreshing allowance; enough to evaluate the model but limited for sustained production use.
For solo designers, marketers, or regular users who need steady day-to-day image generation and want privacy over their prompts and outputs.
For power users and teams producing video content or running Reve as a daily creative workhorse, needing higher volume and multimodal input support.
Genuine layout-control differentiation, but no public funding data and a pricing page split in two.
“Reve's layout endpoints and native 4K output are real technical differentiators. The vendor picture is thin — no team size, no funding, no changelog evidence.”
Native 4K generation at $7.99/month Lite tier, with Extract/Create/Render Layout endpoints splitting structure from pixels. That's a genuine feature no one else I looked at, not OpenAI's image models, not Midjourney, not Stable Diffusion, is describing this way. Layout-first editing is a real wedge if it works as documented.
Two questions: who's behind this, and will api.reve.com still resolve in three years? No public funding data, no team size, no changelog. The consumer app and API split (reve.com vs api.reve.com) also suggests a company still figuring out its core business.
The experimental label on the layout endpoints is the tell — Reve itself isn't confident these are stable yet. Pilot the Create/Edit/Remix core with a small team. Don't build critical infrastructure on the layout endpoints until they graduate out of experimental.
Layout endpoints and test_time_scaling are differentiated moves competitors like Midjourney don't publicly offer.
Unproven vendor name next to OpenAI and Stability, but usage-based API pricing limits exposure.
REST API with bearer token auth and versioned models (reve-edit@20250915) suggests fast integration for developers.
Layout-level control and native 4K output advance capability, not just cost savings on existing generation.
No public funding, team size, or time-in-market data that I could find.
Developers who want compositional control over generation and can tolerate an experimental feature set.
You need a vendor with a visible funding history before committing production workflows.
Layout-first composition control that treats structure and pixels as separate design decisions.
“Reve splits layout from rendering, which is the first API I've seen that mirrors how art directors actually compose before they render. That's a real craft advantage, though the ecosystem around it is still thin.”
Extract Layout, Create Layout, Render Layout — that three-endpoint split is the interesting design decision here. Most generation APIs, including Midjourney and Stable Diffusion pipelines, treat composition as something you fight for inside a prompt. Reve treats it as a reusable structural asset, separable from the final render. That's how a layout designer actually thinks, and it's a genuine ceiling-raiser for teams doing systematic, template-driven creative work.
Native 4K (16 megapixels) output and stated resistance to iterative degradation matter more than they sound. Print-adjacent workflows die on upscaling artifacts and edit drift; Reve claims to solve both at the source rather than in postprocessing.
Three-year risk: this is a backend API, not a design tool with a canvas history or version tree, so it slots into your stack as infrastructure, not as a creative surface your designers touch directly. At $7.99 to $19.99 a month, credit-metered pricing is fine for prototyping, less fine once test-time scaling and 4x upscaling start compounding cost per asset.
Midjourney and Stable Diffusion pipelines decide composition inside the prompt; deciding it before the render is a difference a buyer can name.
Extracting a layout, reusing it, then rendering it matches how a layout designer composes before committing to pixels.
Create Layout and Render Layout are clean primitives to wire into a template system, but there is no canvas or version tree for a design team to plug into.
Three years in this sits in the stack as infrastructure, and cost per asset moves with how often test-time scaling and 4x upscaling are switched on.
Separating layout from render is a structural bet on how creative work gets composed, not a feature bolted onto a generation endpoint.
Creative teams who reuse one composition across many renders.
Designers who want a canvas with a version tree they open themselves.
$7.99 to $19.99 sticker. Credit metering is the real bill.
“Two paid tiers, both published. But it's a credit system underneath, and that's where the math gets fuzzy.”
Lite at $7.99/month, Pro at $19.99/month. Both listed, no sales call required. Free tier exists but capped at daily refreshing allowance — fine for evaluation, not production.
Here's the catch: pricing is usage-based credits layered under subscription tiers. Test-time scaling costs extra credits. Upscaling, background removal, effects — all postprocessing, all metered. 50 seats isn't the unit here; API calls are. Year-1 estimate on Pro: $240 base, plus credit overage on video generation (250 video energy/month cap) and 4K renders. Year 3 depends entirely on volume nobody can forecast from the pricing page alone.
Compare to Stability AI's per-image API pricing or OpenAI's token-metered image endpoints — same structural risk, different math. No published overage rate for exceeding energy allocation. That's the number procurement will ask for and won't get from marketing copy alone.
Credits stacked under a subscription hand procurement a unit it cannot convert into a purchase order figure.
Every plan bills monthly, so stopping costs one billing cycle rather than a term to negotiate out of.
Lite at $7.99 and Pro at $19.99 are both posted with no sales call, but the credit rate underneath them is not.
Cost per finished asset is knowable only once volume is known, and the published tiers give no way to estimate it in advance.
Upscaling, background removal and effects are metered separately, which makes the subscription the smallest part of the year-three number.
Growing teams who add headcount faster than they add renders.
Procurement teams who need a published overage rate before signing.
Layout endpoints are a real compositional tool, not a marketing bullet point.
“Reve separates structure from pixels with Extract/Create/Render Layout, which is genuinely useful for anyone who's fought Midjourney's seed roulette. Still API-first, so the daily experience depends entirely on what you build around it.”
Extract Layout, Create Layout, Render Layout — that's a workflow, not a feature checkbox. Separating composition from final render means you can lock a layout and iterate on style without redoing the whole shot. That's the kind of control Midjourney users beg for in Discord threads and never get.
But this is a REST API, not a canvas. No mention of a Figma plugin or drag-and-drop layer panel in the API docs themselves — the consumer editor lives on a separate site. So day-3 reality for a designer means either building tooling or living in the web app instead, and those are different products with different rules.
Remix supports up to 6 reference images addressable by XML tags, which is precise but not visual — you're writing markup to point at image indices instead of clicking thumbnails. 402 for insufficient credits and a content-violation header are good signs someone building this actually hit those errors themselves. Test-time scaling as a quality/cost dial is a nice touch for power users who know when to spend.
By day three the choice is explicit: build tooling over the endpoints, or move the work into the separate web app.
Failure states are spelled out down to the 402 for credits and the content-violation header, but nothing addresses how a designer works without a canvas.
Aiming at a reference image means typing an XML tag around its index, and layout work spends a Create Layout call before anything renders.
Six individually addressable references, a per-request quality dial and pinned model versions like reve-edit@20250915 leave real room to grow into.
Locking a layout and re-rendering only the style fits the revision loop that actually happens, though nothing connects it to a design tool.
Designers who iterate on style without re-rolling the whole shot.
Designers who expect to pick reference images by clicking thumbnails.
Solid API guts, but this is a tool for your dev team, not you
“Reve is a backend API for image generation with real technical depth: layout endpoints, 4K native output, XML-tagged remixing. But there's no app-level polish to judge because most of you will never touch the interface.”
Let's be clear what Reve is. It's a REST API at api.reve.com with bearer tokens and 402 error codes for insufficient credits. That's not a knock, that's just the product. Your developers will live in this, not you.
What's actually interesting: the layout endpoints. Extract Layout, Create Layout, Render Layout — separating composition from pixels is a smart idea, and I haven't seen Midjourney or Stability offer that as a clean primitive. Native 4K generation (16 megapixels) without an upscale pass is also a real number worth noting, not marketing fluff. Test-time scaling is a neat idea too: pay more credits, same latency, better output.
The consumer site exists ($7.99 Lite, $19.99 Pro) but the meat is the API, and for a dev-facing product the docs and error handling matter more than any dashboard animation. The tradeoff here: you're trusting versioned models (reve-edit@20250915) for production stability, which means someone on your team needs to own version pinning, or you'll get quality drift you didn't sign up for.
Multiple response formats and metadata in every response show care, but this is API ergonomics, not UI sweat.
Layout endpoints and XML-tag remixing add real depth for month three, past the basic Create call.
Platform listed as web only; as an API product this matters less, but the consumer site has no stated mobile story.
Bearer token auth and versioned endpoints mean day one is documentation reading, not exploring.
Specific status codes for credits (402) and content violations show someone thought through failure states.
Developers building content or design tools who need programmatic image editing with compositional control.
You want a polished consumer app experience rather than something your engineering team integrates.
Layout endpoints are new. 'Experimental' is doing a lot of work in that sentence.
“Solid API surface with real differentiation in layout control, but docs/no changelog that I could find, though the blog exists. That's a viability gap, not just a marketing one.”
Three tells worth naming. One: a blog, but no docs or API reference I could find, no changelog that I could find — for a dev-facing API product, that's the exact stuff I'd want visible before integrating. Two: 'experimental' on Extract/Create/Render Layout, the actual differentiator versus Stable Diffusion and Midjourney. Three: versioned models like reve-edit@20250915 suggest active iteration, which is a genuine positive signal.
Pricing is honest and specific — $7.99 Lite, $19.99 Pro, native 4K claimed and specific enough to verify. No free trial, though there's a free tier with daily refresh, which is the fair way to gate a credits-based product.
Exit portability is the real question. This is a REST API with bespoke XML-style reference tagging and a proprietary layout format — if Reve folds, you're not just swapping a base URL. You're rebuilding prompt structure. Fine for a 2-year bet. I'd want more public shipping cadence before calling it a 3-year one.
Layout-level compositional control before rendering is a real gap versus Midjourney and Stable Diffusion's prompt-only workflows.
Proprietary XML-tag reference syntax and layout format mean migration to Stability or OpenAI's image API isn't a drop-in swap.
A blog exists, but no docs or changelog that I could find — hard to gauge shipping cadence or support depth from what's public.
Claims (native 4K, zero-degradation edits) are specific and testable, not vague superlatives.
Versioned model releases (reve-edit@20250915) match healthy iteration patterns, but experimental-tagged core features is a yellow flag.
Developers who need programmatic layout control and can tolerate experimental endpoints.
You need a stable, well-documented API today with a clear migration path off it.
Common questions answered by our AI research team
Reve 2.1 generates images natively at 4K x 4K resolution, true 16 megapixels, making them suitable for high-quality print workflows right out of the box without a separate upscaling step.
Yes. Reve provides API endpoints to edit existing images using text instructions, in addition to generating images from text.
Yes. Reve's API supports remixing images using reference images combined with prompts as part of its core endpoints.
Reve 2.1 is designed to resist degradation. Image edits with references remain more stable through longer iterative workflows, and generating images without references from code results in zero degradation with no accumulation of artifacts.
The experimental layout endpoints let developers extract, create, and render structured image layouts, giving precise compositional control over an image before it is rendered.
Who Reve competes with, and how each one differs
Midjourney runs on a Discord-native, prompt-only workflow with the largest mainstream user base of any image model, while Reve's layout-representation architecture lets users directly reposition objects and edit text inside an image after generation.
Black Forest Labs ships FLUX as open-weight models that developers can self-host or fine-tune, versus Reve's closed, hosted platform built around its proprietary layout-editing UI and API.
Ideogram competes closely on legible in-image text and prompt adherence, but lacks Reve's structured 'layout' representation that lets users move, resize, and re-edit individual objects after generation.
Imagen is distributed as a feature inside Google's Gemini apps and API rather than sold as a standalone creative product, while Reve is a dedicated professional image-creation and editing platform with a 4K-focused, layout-based editor.
Recraft is built around vector graphics and brand/design-system templates for marketing teams, whereas Reve focuses on photorealistic, layout-based image generation and editing at native 4K.
OpenAI's GPT-4o/image tools are bundled into the general-purpose ChatGPT assistant rather than a dedicated creative-editing product; notably OpenAI also became an investor in Reve in July 2026 even as some of Reve's research staff moved to OpenAI, an unusual competitor-investor relationship.
Reve is an AI image generation and editing platform based in Palo Alto, California, combining natural-language prompts with a drag-and-drop editor.
