Tasting Notes: Where the FLORA team reviews the latest text, image, and video models

Recraft V4 Doesn't Look AI-Generated. That's the Point.

Recraft V4 is a text-to-image model with a strong editorial sensibility. It produces images that feel grounded in real photographic tradition, the kind of output that reads as intentional rather than generated. This review of recraft is written for creatives and creative teams evaluating how and where the model belongs in a production workflow.

Recraft Is Changing What's Possible in Graphic Design

For teams working inside node-based workflows, the question is never just whether a model produces good images. It's whether those images are useful at the specific moment in a workflow where they're needed. Recraft V4 earns its place because its output is consistently close to production-ready. Less correction, less iteration, and more time spent on the work that actually requires creative judgment.

Editorial Photography inside FLORA

Recraft V4 has a distinct visual character. The output carries a Leica and Fujifilm quality to it, with grain, texture, and tonal depth that feels closer to film than most models produce. Environments and characters read as diverse and considered rather than generic. The result is work that feels like it came from a creative direction, not a prompt.

• Film-influenced texture and grain out of the box

• Environments and characters feel diverse and editorially considered

• Accurate rendering of traditional mediums including charcoal, graphite, and watercolor

• Noticeably less stock-image quality than comparable models like Nano Banana Pro and Grok

• Output has a natural, refined feel that holds up in professional creative contexts

Best used in workflows where the hero image needs to land without heavy iteration.

Commercial Product Photography inside FLORA

Clean, brand-ready output with minimal correction needed downstream. The model understands the conventions of commercial photography well enough that what comes out of the node is close to usable on first pass.

• Accurate proportions, clean backgrounds, no artifacts

• Holds the aesthetic logic of a product shot including negative space, lighting, and surface

• Moves directly into design systems without significant post-processing

Best used in workflows where output feeds directly into a layout or brand system.

Atmospheric and Scene-Setting Work inside FLORA

For work that needs to establish a specific emotional register, the model is consistent. Mood translates reliably across variations, which matters when you're generating at volume inside a workflow.

• Tonal consistency across batch outputs

• Scene elements work together rather than competing

• Feels considered, not assembled

Best used in workflows where multiple variations need to hold the same tone across a campaign or layout system.

Isolated Text and Typographic Elements inside FLORA

Stronger than most models at rendering text as a visual element when the context is clean and the text is singular. Useful for design nodes where type needs to feel embedded in an image rather than composited on top.

• Correctly spelled, well-formed output for signage, handwritten notes, and woven text

• Handles material integration like neon, fabric, and handwriting without breaking legibility

• Feels native to the image, not added

Best used when a single typographic element needs to live inside the image. Avoid for dense or multi-element text.

Where the Model Has Limits

Knowing where Recraft V4 breaks down is as useful as knowing where it excels. These are patterns that surfaced consistently across testing, not edge cases.

• Complex human dynamics. Scenes that depend on believable interpersonal relationships or gestural accuracy between figures don't land well. The model can render people, but it struggles with the logic of how people relate to each other in space.

• Image-to-image transformation. When given a reference image and asked to transform its style while preserving its structure, the model tends to reinterpret rather than transform. Object placement shifts and architectural logic loosens. Use it for generation, not structural transformation.

• Dense or integrated text. Product labels, packaging, multi-element interfaces. The visual rendering can appear fine while the actual copy is wrong. Not reliable for any work where text accuracy inside the image matters.

• Niche or historically specific contexts. The further a prompt moves from well-represented visual territory, the less predictable the output becomes. Medieval settings, specific scientific environments, complex cultural contexts are all areas where results become inconsistent.

Recraft v4 Prompt Pack

Try the model with these prompts below.

Recraft V4

Best for: high-quality general images with enhanced detail

A moody coffee shop interior at golden hour, steam rising from espresso cups, warm bokeh lights, film grain texture, intimate and cozy atmosphere

Recraft V4 Pro

Best for: complex detailed scenes, premium quality

Portrait: A close-up portrait of a weathered fisherman, deeply textured skin, salt-sprayed jacket, fog rolling in behind him, shot on medium format film

Product: A luxury perfume bottle on a marble surface, surrounded by dried botanicals and amber light, editorial product photography aesthetic

Recraft V4 Vector

Best for: clean vector illustrations, icons, UI assets

A set of four weather icons — sun, cloud, rain, and snow — in a clean modern vector style, bold outlines, flat fills, suitable for a mobile app UI

Recraft V4 Pro Vector

Best for: detailed vector illustration, brand art, poster graphics

A bold retro concert poster for a jazz festival — trumpet player silhouetted against a geometric sunburst, vintage color palette of burnt orange, cream, and deep navy, intricate linework