LoginSign Up
← Back to all sparks
R

Recraft

DESIGN
Velocity7.5

Recraft bets on brand consistency and an agentic studio as AI image generation commoditizes.

ai-imagebrand-consistencyagentic-aimulti-modaldesign-platform
◆Current state
Recraft Studio has shipped at an accelerating pace since August: ChatGPT Images 2.5 at 50% lower latency, V4 Styles (consistent visual style across all generations without model training), voice input, multi-model side-by-side comparison with batch sizes up to 8, a full interface redesign introducing a Chat (agentic) mode, and Seedance 2.5 for video clips up to 30 seconds. The product has expanded from image generation into a multi-modal creative suite with video, mockup, and set-of-images generation types.
◆Where it's heading
V4 Styles and the Design Agent (currently in open testing) are the clearest signals: Recraft is moving toward brand-consistent asset generation at scale, not one-off creative exploration. The Chat interface giving users an agentic control layer—describe a product, get a full design kit—positions Recraft as a tool for maintaining visual identity across a run of assets, not a sandbox for inspiration. This is a move toward brand infrastructure, not creative tools.
◆Prediction
The Design Agent exits testing and launches to paid subscribers within the quarter; expect V4 Styles to be the anchor use case in e-commerce marketing, where product visual consistency across hundreds of SKUs is a direct revenue lever.

◆Recent moves

  1. 4d ago

    New Model: ChatGPT Images 2.5

    OpenAI's ChatGPT Images 2.5 model is now available in Recraft Studio, delivering the same output quality as its predecessor at 50% lower latency. For high-volume generation pipelines where speed matters as much as quality, this is a practical improvement—faster iteration without the usual tradeoff of accepting worse outputs.

  2. 20d ago

    V4 Styles

    ⚡ SPARK

    V4 Styles lets users define a visual style once—from a library pick or their own reference—and every subsequent generation holds to it without model training or per-image prompting. This is less a feature than a constraint system: Recraft is now a tool for maintaining brand identity across a run of images, not just generating individual ones.

  3. 22d ago

    Voice Input

    Voice input lets users speak their prompt rather than type it—most useful on mobile or when iterating rapidly across multiple concepts. It reduces the friction of the generation loop without changing what the product can produce.

  4. 23d ago

    Multi-Generation & Model Comparison

    Multi-generation batches (1, 2, 4, 6, or 8 images) and side-by-side model comparison let users run more experiments per prompt. These are workflow accelerators—the kind of feature that makes Recraft stickier for power users testing model quality, without changing the product's fundamental positioning.

  5. 25d ago

    Recraft Studio's new interface

    ⚡ SPARK

    The new Recraft Studio interface introduces a Chat (agentic) mode alongside a restructured panel layout, and adds generation types beyond images: Vector, Video, Mockup, and Set of Images. The interface redesign isn't cosmetic—it reflects a product that now does materially more than it did six months ago and needed a new information architecture to contain it.

  6. 1mo ago

    New Model: Seedance 2.5

    Seedance 2.5 for video arrives in Recraft Studio, supporting clips up to 30 seconds at 480p (18 credits) or 720p (41 credits). Extending from ~6-second loops to 30-second scenes makes AI video generation practical for social media and short-form content—a meaningful step up even if the model itself is sourced externally.