AI Weekly Update: Six leaps that just turbo-charged the creator’s toolbox
Jaroslav Urbánek, founder of TECHNOMATON 4 August 2025 2 min read
This article was published on and describes the situation as of that date.
TL;DR
Six releases in seven days: faithful character cloning, a “thinking” LLM, an in-chat tutor, desktop-grade video gen, photo-real TTI, and whole virtual worlds at the push of a prompt. The AI train isn’t slowing down; it just blew past the next station without tapping the brakes.
1. One photo, infinite clones
Toronto-based Ideogram rolled out Character, a model that keeps a character on-model across any prompt with nothing more than a single reference shot. Think of it as owning one LEGO minifig and instantly dropping it into every new play-set you dream up. The feature is live on the web and the iOS app and, for now, it’s totally free.
2. Gemini 2.5 switches on “Deep Think”
Google’s Ultra tier just gained a mode that lets the model linger, juggle several hypotheses in parallel, and only then answer. Picture a chess grandmaster silently exploring half a dozen lines before moving a pawn. In early tests the upgrade aced LiveCodeBench V6 and earned bragging rights at this year’s Math Olympiad trials.
3. ChatGPT turns tutor
OpenAI’s new Study Mode refuses to hand over the solution on a silver platter. Instead, it nudges you with questions, hints, and lightning-fast quizzes until the lightbulb flips on. It’s already available in the free tier, so the “I don’t know how to study” excuse just ran out of road.
4. Full-HD video from text on a single RTX 4090
Alibaba’s open-sourced Wan 2.2 cranks out a five-second 1080p clip in under nine minutes - even on consumer hardware. With a MoE backbone, a LoRA style slider, and extras like volumetric light, it feels less like a camera and more like a pocket-sized movie studio.
5. FLUX.1 Krea: photos minus the “AI sheen”
A collaboration between Black Forest Labs and Krea produced a text-to-image model that finally ditches the waxy skin and neon hues. The renders land closer to what you’d expect from a pro wielding a medium-format camera on a cloudy day.
6. HunyuanWorld 1.0: prompt → explorable 3-D world
Tencent’s brand-new pipeline spins a single sentence - or even one photo - into a layered, mesh-exportable 3-D environment you can stroll through. Imagine sketching a treasure map at breakfast and walking its beaches by lunch. VR studios, take note.