Ask five different creative teams which AI image tool is the best, and you’ll likely get five different answers, because they’re each testing the tools on different jobs. Concept exploration, retouching, branding work, and mood boarding pull on different strengths, which is why our stack includes more than one tool. At SketchDeck, our designers test these tools on our own brand before we ever bring them into client work, and what we’ve learned is that the “best” AI image tool depends entirely on what stage of the creative process you’re in.
This is the first post in a series where our AI Exploration Team shares hands-on ratings and firsthand takeaways on the AI tools currently in rotation at SketchDeck. Each tool is scored on a simple five-star scale based on real use, not marketing claims. Here, we’re focused on concept and image generation: the tools we reach for when we’re exploring ideas, building mood boards, generating hero imagery, and finishing final assets.
Midjourney — ★★★★★
The Default for Concept Exploration
When a project starts with a blank page, Midjourney is usually where we go first. It produces the strongest aesthetic quality of any image generator we’ve tested, and its style and character reference features let us keep visual consistency across a full series of images, which matters when a client needs more than one hero shot.
Where Midjourney falls short is accuracy. Text, hands, and branded products still need manual cleanup, and without careful prompting, output can drift into a look that reads as distinctly “AI-generated” rather than on-brand. For commercial work, we always run the paid tier to stay within usage terms.
Bottom line: Our default first tool for concept work and stock photo alternatives.
Nano Banana — ★★★★☆
Editing Where the Design Already Lives
Nano banana solves a specific problem well: making an existing image work inside a layout without leaving Figma. Inpainting, object swaps, and background changes happen right where the design is being built, which removes the tool-switching that slows down iteration.
It has fewer controls than running the underlying model directly, and it isn’t a substitute for Photoshop when precision matters. It also carries a watermark when used directly on the Gemini web interface.
Bottom line: Our default for image edits during the design phase itself.
Adobe Firefly / Photoshop AI — ★★★☆☆
Built for Finishing, Not Originating
The strongest argument for Firefly has less to do with aesthetics and more to do with licensing. Because it’s trained on licensed Adobe Stock content, it comes with commercial indemnity, which is often the deciding factor for enterprise clients who need to know exactly where their imagery rights come from. Combined with zero friction inside an existing Photoshop workflow, Generative Fill, Generative Expand, retouching, and object removal are all faster with Firefly in the mix.
The tradeoff is aesthetic strength. Firefly’s output is weaker than Midjourney or nano banana when it’s used as a primary generator, so we treat it as a finishing tool rather than a starting point for hero imagery.
Bottom line: Indispensable for post-production. Not where we go to originate a concept.
Recraft — ★★★☆☆
Native Vectors for Branding Work
Recraft is the only major tool in our stack that generates true native vector output instead of raster images, which makes it worth keeping around for branding-specific needs like posters, logos, and layouts that require structured negative space for typography.
That said, results swing widely from prompt to prompt, and the polished examples you see in marketing materials don’t reflect what we consistently get in practice. We rarely use Recraft output as client-ready without a round of manual refinement.
Bottom line: Worth having for specific vector needs, with close supervision.
Google Mixboard — ★★☆☆☆
A Playground for Early Concepts
Mixboard is built for AI-powered mood boards on an infinite canvas, and it’s faster than traditional Pinterest-style reference hunting. Describe a direction in natural language and get visuals back, then use the combine feature to blend styles between images. We reach for it in the earliest concept rounds, especially when the references we need don’t exist yet and have to be invented.
As an experimental Labs tool, features are still settling, and it offers less control than working directly in Midjourney once craft starts to matter.
Bottom line: A strong concept-stage playground. Midjourney still wins once output needs to be production-grade.
What This Round of Testing Told Us
Five tools, five different jobs: Midjourney for the first concept pass, nano banana for edits without leaving Figma, Firefly for cleanup after the fact, Recraft when a client needs an actual vector file, Mixboard when the references don’t exist yet. Picking the right one for the job in front of us, rather than defaulting to whichever tool is newest, is what our AI Exploration Team spends its time figuring out. Next up in this series: the tools we use for video, voice, and motion.
Partner with SketchDeck
Matching the right tool to the right stage of a project, and knowing where AI’s job ends and a designer’s begins, is exactly the kind of thinking our team brings to every project.
Curious how this could work for your brand? Let’s talk.
