Recraft V4 Styles

Recraft V4 Styles is a dedicated image model for generating different subjects under a shared visual reference. It is worth evaluating when a campaign needs a recognizable illustration or photography style across several assets. The decision is whether the look remains useful as the subject, crop and composition change. This source-based evaluation proposes a campaign test; no generated set has been inspected by tasarim.ai for this article.

Define the visual rules before generating a set

A recognizable campaign style is more specific than “warm” or “minimal.” It may depend on a restricted palette, rough edges, shallow shadows and a particular relationship between subject and empty space. Write those observable rules before selecting references. Otherwise, it becomes easy to accept whichever image looks appealing, even when the set no longer belongs together.

Consider a fictional public-library campaign with images for a reading club, a repair workshop and a neighborhood seed exchange. The subjects differ, but the printed and digital materials should feel related. A reference might establish cut-paper shapes and muted colors without prescribing books in every scene. That distinction matters because style consistency should not force every asset to repeat the reference's subject.

Recraft introduced V4 Styles in August 2026 as a reference-based model line. Its public explanation describes using references without a separate training step. That makes it a candidate for testing an existing art direction, rather than evidence that the system has learned your full brand rules. Keep the approved logo, typography and factual campaign text under separate control until they have their own acceptance evidence. Recraft announcement

Identify Styles separately from V4.1 and Studio

The standard raster model is recraftv4_styles. Styles Pro, Styles Vector and Styles Pro Vector use separate identifiers. Raster output, editable SVG and a higher-resolution Pro result are not interchangeable promises. Record the selected identity when comparing exports, particularly if an image will eventually be placed in a large printed layout. Styles API model page

Recraft V4.1 is the current core image generation line and the general API default. Styles remains a dedicated line, with reference-conditioned requests following a different default. This is a reason to specify the model explicitly in a technical test. An omitted field can turn a comparison into a test of default routing rather than of the intended model. V4.1 API documentation

Recraft Studio is the commercial interface, not another name for the model. Its current documentation says the combined V4.1 selector includes Styles. If the UI does not expose the resolved model for a particular output, record that limitation instead of assuming API-level reproducibility. A saved style in the product and an exported image are also different artifacts: one conditions later generation; the other is material you inspect and place in a design. Studio availability

Assemble references that describe one visual idea

For the library experiment, begin with approved assets whose rights allow their use as references. Choose images that agree on the properties you want to preserve. A textured paper illustration and a glossy product photograph may share colors while disagreeing about rendering. Combining them without a clear intention makes it difficult to explain why a result should count as correct.

The documented API accepts PNG, JPG and WEBP references, up to ten images with a combined size limit and per-file limit. It supports a saved style identifier or attached references, but rejects a request that supplies both. Check the current limits before preparing a large reference board. A board full of tiny samples can technically be one image while remaining a poor evaluation input because its useful details are hard to inspect. Input and endpoint documentation

Make a small reference record describing why each selected image belongs. For example, one controls edge texture and another shows how people are simplified. If the images disagree about shadows, choose which direction governs the campaign before generating. This is the article's proposed preparation method, not a claim that the API exposes independent controls for every visual property.

Test subject changes with a contact sheet

The following workflow is unexecuted. Prepare three prompts for the library campaign: a reader beside an open book, someone repairing a small appliance, and hands exchanging seedlings. Keep the intended palette, framing and medium consistent. The point is to ask for genuinely different content while giving the style a fair opportunity to hold.

Create the style once and record its identifier, reference set, selected model, match mode, output settings and date. Recraft documents a reusable style creation step followed by image generation. Reusing a known style gives the test a stable reference; it does not guarantee identical results across future service changes. Save the source files and request metadata in the project's approved storage location. Reusable style workflow

Place the resulting images beside the reference in a contact sheet at the sizes they will occupy in the campaign. First compare the set without reading the prompts. Then inspect whether each image communicates the intended activity. A collection can have consistent texture while failing to distinguish a repair event from a reading group. Both checks are needed before deciding that style control solved the campaign problem.

Make a second row with the same prompts and a deliberately different composition requirement, such as more empty space for a heading. This reveals whether the style survives a practical layout constraint. Do not report this as an executed experiment until the exact inputs, returned images and reviewer observations have been recorded.

Compare precise matching with useful flexibility

Recraft documents precise and flexible style matching. Those are controls to evaluate against a brief, not automatic quality labels. For the library campaign, precise matching may be useful when all three posters share a tightly specified visual language. A more flexible result may fit a campaign whose different events need more varied compositions. The correct choice depends on which reference properties are mandatory. Match modes

Review palette, edge treatment, texture, lighting and placement separately. Mark a difference as acceptable only when the brief permits it. A seed-exchange image can contain green plants while still belonging to a muted campaign; requiring every asset to have identical color proportions may suppress the subject. Conversely, introducing a glossy highlight into a matte paper style may matter even when the overall colors look close.

Ask the reviewer to identify the outlier without seeing its model or mode label. If the choice changes when labels are revealed, keep the blind observation in the record. This small procedure limits the influence of expectations about a newer model. It is not a statistical benchmark and should not produce a claimed win rate from a handful of campaign images.

Inspect the intended export, not only the preview thumbnail. If a design needs editable vector shapes, test the separate Vector model and open its SVG in the target editor. If the task needs a larger raster file, evaluate the separate Pro identity. Neither choice removes the need to inspect shapes, crop, readability and the final layout.

Separate generation price from production readiness

The current model-specific API table lists standard Styles raster generation at $0.035 per image and style creation at $0.005 per request, checked September 12, 2026. The broader pricing page lists separate rates for Vector and Pro variants. Treat these as dated service charges, not a complete project budget or an app-subscription allowance. Styles API pricing, pricing page

Estimate a campaign by counting planned attempts, selected variants and review work. If a style will be reused, distinguish creating it from generating another image with the existing identifier. Do not budget every attached-reference request as if it were a free lookup. In a small project, manual corrections and art-direction review may determine the total effort more than the generation line item; that is a planning consideration, not measured cost data from this evaluation.

Web subscriptions and API access have different terms. Recraft's subscription guidance distinguishes free-plan assets from paid-plan commercial ownership. Developer terms govern API outputs, input permissions, storage and service restrictions. Confirm the applicable route before placing client material into the workflow. Having the right to use an output does not establish rights to someone else's reference image or permit training another model with the asset. Commercial rights guidance, Developer Terms

Repair the cause of a style mismatch

If all outputs inherit an unwanted object from the reference, the reference may be communicating content too strongly for the test. Try an approved reference that expresses the same texture and palette with a different subject. Keep the prompt and model fixed so the next result can reveal whether the reference choice mattered. This is a proposed troubleshooting action, not a proven cure for a documented model defect.

If the style holds but the scene becomes unreadable, simplify the subject arrangement before changing every reference. A repair-workshop poster needs recognizable hands, tools and activity at its final display size. The most faithful texture is of little use if the viewer cannot tell what the event is. Use manual illustration when the required visual explanation remains too specific for the generated candidates.

If the API rejects the request, inspect identity, reference presence, file limits and incompatible fields. A no-style error belongs to request preparation, not to image quality. For inconsistent behavior after a previously usable setup, compare the saved request and current documentation before assuming the brand reference has stopped working. Record what changed and preserve the older accepted asset until the replacement has been reviewed.

Decide what the evidence supports

Recraft's documentation describes a company-designed style benchmark with outside blind evaluators, dated August 14, 2026. That is not the same as an independently designed evaluation, and the full methodology is available only on request. This article does not convert that report into an overall quality ranking or repeat its percentage as a promise for your campaign. Benchmark scope

For a single image with no stable reference direction, ordinary image generation may be sufficient. For a reusable set, Styles is a defensible model candidate because its input and reuse mechanism directly match the task. For exact typography, regulated wording or a registered logo, keep those elements in controlled design tools and assess the generated artwork as one component of the composition.

The evaluation was prepared with AI assistance from first-party sources on September 12, 2026. No independent campaign outputs, measured latency or acceptance rate exist for this draft. The next useful step is the library contact-sheet trial, followed by a designer's explicit decision on reference selection and manual correction. Accept the workflow only for the visual rules it actually preserves under that test.