The blank prompt box has long been the creativity killer of mainstream AI tools, because it asks newcomers to conjure specificity on command while rewarding insiders who learned obscure phrasing hacks that nudge models into coherence. Into that gap arrives Mixup, an iOS app from Things, Inc. that reframes image generation as a guided, social game built around “recipes” rather than freeform prompts. Instead of typing a wish and hoping the model understands, users start with Mad Libs-style templates, drop in personal photos, text, or quick doodles, and receive outputs paired with the exact recipe that produced them. That pairing matters: it makes the creative process legible, repeatable, and shareable. Early adopters can browse a public feed, tap Try recipe, and apply a template to their own material, turning the prompt itself into a reusable artifact that circulates like a meme with provenance.
A template-first approach to generative imaging
Mixup’s core mechanic hinges on Google’s Nano Banana image model, which the team says handles likeness in a convincing but non-creepy way—critical for an app that invites users to remix photos of themselves and friends. Templates reduce the system’s “slot machine” feel by anchoring generation to structured inputs, while before/after views (when enabled) provide an immediate reality check on what changed. That transparency addresses a persistent pain point for casual creators: unpredictability that forces endless retries or prompts that get longer with every fix. With Mixup, the context of each output is part of the content, so anyone can audit the steps, reuse them, or tweak them without guessing. The result is a loop that feels closer to collaborative editing than occult spellcasting, with recipes functioning as creative scaffolding rather than cages.
The app treats prompts as media objects—and that social framing matters as much as any model benchmark. Users can publish results alongside their recipes, follow creators whose styles resonate, and grant permission for “mixables,” a consent-based option that lets followers generate with a person’s likeness. That mechanic resembles the Sora ethos of starting with source material, but it wraps the experience in norms: opt-in identity sharing, visible recipe lineage, and a feed where successful templates become reference points for others. Sketching inside the app lowers the barrier even further; doodles and captions can serve as inputs when a polished photo isn’t available. In practice, this shifts the learning curve from prompt engineering to playful experimentation, where control comes from template selection and small edits instead of verbose incantations.
Business model, safety, and what comes next
Mixup opens with 100 credits for new users—valued around $4—with per-image costs landing near four cents and paid tiers offering 100, 250, or 500 monthly credits. That pricing anchors the experience in lightweight, repeatable creation rather than sporadic big-ticket sessions, while the invite gate at launch helps the team meter demand and tune moderation. On-device performance is tuned for iOS 26 and supported from iOS 18 upward, signaling a mobile-first bet where instant feedback and touch-native controls keep the loop tight. Safety layers combine model-level restrictions from Google with OpenAI-powered tools to filter sexual content and violence, an approach that balances speed with policy coverage. The claim that Nano Banana preserves identity convincingly underpins likeness-based features, but the product design adds guardrails through consent and recipe transparency that make social reuse feel grounded, not exploitative.
The broader trend driving Mixup had been a pivot from raw horsepower to product craft: guided creativity, shareable workflows, and social norms that make AI usable by people who will never write a baroque prompt. Templates, before/after context, and remixable recipes reframed generation as a participatory pastime rather than a solitary grind. If adoption grew, the team could have extended the model to web or Android and formalized a marketplace where recipes, aesthetics, and opt-in likenesses circulated as creative currency. For now, the strongest next steps rested on hardening moderation at scale, surfacing attribution for derivative recipes, and refining credit economics to keep everyday creation affordable. In that balance of control, safety, and play, Mixup demonstrated how design—not just models—could unlock a durable, creator-friendly ecosystem.
