Red Panda. Powerful AI Image Generator by Recraft

A new powerhouse has arrived in AI image generation, stirring up excitement among fans and pros. Meet 'Red Panda', the latest model climbing the ranks and outshining industry names like Flux 1.1 Pro and MidJourney v6.1.

Developed by the team behind Recraft V3, a pro-level tool for designers, Red Panda is built to impress. Let's explore what sets Red Panda apart and why it’s drawing in the AI world.

In the past few days, Red Panda has dominated the Artificial Analysis leaderboard, a ranking system based on user choices on randomized prompts. This leaderboard isn’t just about expert opinions; it's driven by user choices, giving Red Panda the top spot among AI image generators. In the Text-to-Image Arena, users are shown two anonymous images generated from the same prompt and asked to pick their favorite based on quality, lack of artifacts, and prompt accuracy. Red Panda’s 72% win rate and high ELO score of 1172 place it 40 points ahead of its closest rival, FLUX 1.1 Pro by Black Forest Labs, known for powering the Grok image generator.

In a twist that feels like a Scooby-Doo mystery, Red Panda is actually Recraft AI, a UK-based platform offering various creative features. Besides its image generation, Recraft includes tools for background removal, color palette selection, and both in-and-out painting.

Standout Features of Recraft's Image Model

1. Diverse Styles and Text Capabilities
   Recraft AI offers multiple styles, including raw, photo-realistic, and vector art. But one of its standout features is its ability to generate images with extended text. While many platforms struggle to create visuals with more than a few words, Recraft V3 excels at this, making it ideal for detailed designs like 80s VHS box art or sci-fi movie posters.
2. Accurate Object Counting and Placement
   Recraft V3 also shines in handling prompts with specific object counts, colors, and positioning. For example, a prompt like “Three apples, two pears, and six plums are in the basket” will be accurately turned into the image—a huge win over other platforms that falter on detailed prompts especially where counting objects is required.

Notable Features of Recraft's Platform

Recraft offers a suite of AI-driven tools tailored for designers, including an AI Image Generator for creating visuals from text, and an AI Image Vectorizer to convert raster images into scalable vectors. It also includes an AI Vector Generator for producing vector graphics from descriptions, ideal for logos and icons, alongside a Mockup Generator for visualizing designs in various contexts. Other features include an Image Upscaler to enhance resolution, a Background Remover for isolating subjects, and an AI Eraser for removing unwanted elements, all enhancing control and precision in the design process.

Multiple Image Formats Recraft lets you export your images as SVG, PNG, JPEG or even Lottie files.

Recraft's Image Prompt Examples (test)

OK but let's put this to the test and generate some images with our 50 free credits today.

So here are some crazy Red Panda prompt examples:

The text was a bit confusing, but hey, only a couple letters mistake, that's still pretty good, the text remains readable.

You can see that by default your image generator appears to be part of a potential workflow, ready to be used with other tools offered by the platform. There is toolbar at the top with more options. This is a complete studio, not just an image generator.

POV shot example in Red Panda (Recraft)

This next one was a 'Halloweenize' toggle on:

This isn't bad but it was a killer one in Flux selfies, in my opinion.

And they're not 5, they're 3 dogs. So the claim about knowing how to count precisely is not necessarily strictly true. They're all good dogs though! 

Humans though I can see still might have some problems with their eyes. Not exclusive to Red Panda this issue, just pointing out it's still there. But let's try again with a closer view, usually is much better.

This came out nicely. Let's test their 'Remove background' tool on this image.

And that is really cool, ladies and gentlemen. Obviously, the red background is mine, I put a contrasting color there to highlight the outlines of the woman. But initially I was able to download it as pdf with a fully transparent background. You could pick at that bit at the back of her pony tail which looks like a cloud or smoke, which could be more defined, but overall it's great. 

Another test prompt for anatomy and text again:

'Solid blue background' was turned into a realistic setting, and she's not holding the sign, it's hanging there. But it is a complicated prompt, like have you seen many acrobats holding banners? So fair enough Text is almost right.

With all such powerful tools in one platform I can see how Recraft may attract many AI artists, not sure about pro designers, but there's a whole bunch of people who are interested in tools helping them manipulate AI generated images. I think they have a potential to fill at least some of that niche. Assuming this is just a start and they'll keep improving their model. I mean, it's good already, but I'm not blown away with the results. They're  competitive with the existing models.

Recraft's Pricing and Accessibility

Recraft AI has a free tier offering 50 daily credits, good for up to 25 images per day. Paid plans start at $10, $24, and $48, keeping it accessible for casual users and pros alike. For free plan users, note that image ownership isn’t included.

Generation or modification of a single image usually costs 1-2 credits.

Recraft also offers API access, documentation available at https://www.recraft.ai/docs#features Currently it seems like $10 would buy you 10000 credits.

Redditors Feedback on Recraft

Redditors had mixed reactions to the new Recraft model, with much of the discussion centering around the model's closed-source nature, prompting disappointment among those hoping for an open, local solution. Many users expressed a lack of interest upon learning it was not available for self-hosting, which is a preferred feature in the AI community. Others took issue with Recraft’s website, mocking its outdated design, reminiscent of early internet aesthetics. The lack of visual polish on the site—such as inconsistent kerning in banners and a “geocities vibe”—seemed ironic to users, given that this was meant to be a model targeting professional designers. Several Redditors with design backgrounds were especially critical, calling it a “visual war crime” and noting the disparity between Recraft’s claimed audience and its execution.

The model itself received more favorable, if cautious, feedback for its technical capabilities. Users noted that while Recraft delivered impressive prompt adherence and displayed good anatomical consistency, it struggled with certain aesthetic details, especially in faces and hands. Some found its outputs on par with or slightly behind Flux and other leading models in terms of fidelity. One user highlighted its strong SVG generation capabilities, while another observed the coherence of Recraft’s style, describing it as more rigid than Flux but arguably better suited to simplified design workflows. Despite some positive experiences, the closed-source nature and high subscription costs made many question its value over existing open models.

Some users found it amusing that the high ELO ranking of the model might be skewed by selective voting, suspecting that the model might have “gamed” its way to a top position. This led to critiques of the broader “arena” ranking system and a few back-and-forths about which models delivered the best value. While a few users saw potential in the model for specialized design tasks, many remained skeptical, pointing out that Recraft’s limited flexibility and aesthetic shortcomings ultimately hindered its appeal. You can read the discussion yourself at https://www.reddit.com/r/StableDiffusion/comments/1gfvino/red_panda_is_recraftai/

Last modified 01 February 2025 at 17:50