Luka Kids has introduced an AI drawing projector at IFA in Berlin that treats a child's voice as the beginning of a picture, not the end. The device listens to a spoken prompt and casts a rough outline onto paper. From there, the child traces, colours and elaborates. It is an unusual formula in an age of generative AI that can produce entire scenes in seconds, and it may become an important model for AI-enabled toys.
The projector is built around a deliberately narrow task. A child might say, for example, 'a cat in a spaceship' or 'a castle guarded by dinosaurs.' The device then projects a simple outline of that idea. It does not render a finished illustration. It does not add colours, textures or backgrounds. The system creates a starting point and stops.
Luka says the projector will arrive with 384 templates. The company has not announced pricing and expects the device to be available later this year. It is being shown at IFA, where consumer technology brands often compete to demonstrate how much automation they can pack into a product. Luka is taking the opposite approach by giving the youngest users the most important job: finishing the artwork.
Why stopping is the key feature
The choice to pause after the outline is significant. In a market where children are adopting generative AI tools at a much faster rate than adults, many toys now try to do as much as possible for a child. Some devices generate stories, answer questions, complete homework exercises or hold extended conversations. Luka's projector does not follow that pattern. It collects one spoken prompt, projects one outline and then removes itself from the creative process.
That design choice may be partly about child development. Drawing, colouring and adding details help children practice fine motor skills, visual imagination and narrative thinking. If an AI device draws the whole picture, it may provide a prompt but replace the act of creation. Luka's device is closer to a digital stencil. It gives children an idea to hold onto, but it requires them to make marks on the page, choose colors and decide what happens next.
The product also avoids the social and conversational layer that has created headaches for other AI toy makers. There is no chat window, no persistent back-and-forth and no long-running relationship with a virtual assistant. A child speaks, draws, and moves on. That limited interaction may make the projector easier for parents to supervise and easier for regulators to assess.
A family of AI devices aimed at children
Luka Kids is not new to children's technology. The company's reading robot has been upgraded with large language models so that it can read unfamiliar picture books aloud and answer questions about a page. It works in English, Spanish, German and Chinese, giving families a multilingual reading companion that does not need the book to be from a closed catalogue. The robot can interpret a page on the fly, which is a more complex task than playing a preloaded story.
The company also makes an AI Cube, a camera-equipped device that uses multimodal AI to identify objects around a child and answer questions about them. The cube is designed to turn ordinary rooms into learning spaces. It lets children point at something, ask what it is and receive an explanation. That product is already on sale in Asia-Pacific markets, showing that Luka has begun shipping AI toys beyond the experimental stage.
The new drawing projector extends this portfolio in a different direction. Instead of answering questions or reading aloud, it responds to a creative request. It is less like a tutor and more like an art partner that knows when to stay quiet. The child takes the drawing from an idea to a finished page, and the device does not hang around after the outline has been projected.
Europe tightens the rules for AI toys
Showing AI toys in Berlin also means facing a new regulatory climate in Europe. The European Union adopted a new toy safety regulation in December, and it applies to connected products in ways that the old rules did not. Manufacturers must assess health risks associated with digitally connected toys, including mental health considerations. Products that use AI must also comply with the EU AI Act, which creates obligations around transparency, data governance and risk management.
The mental health dimension is a notable addition. Traditional toy safety standards focus on physical hazards such as choking, sharp edges and toxic materials. The new rules recognise that AI-enabled toys can affect a child's emotional state, development and sense of safety. A toy that responds unpredictably, repeats inappropriate content or encourages compulsive use may be considered unsafe for reasons that have nothing to do with its plastic parts.
The regulation also changes the compliance paperwork. The previous declaration of conformity will be replaced by a digital product passport, which must be accessible through a QR code and kept for ten years. That means manufacturers cannot simply tick a box at design time. They must maintain records that explain how the toy works, what data it collects, how AI decisions are made and what safeguards are in place. The 2009 Toy Safety Directive will be repealed on 1 August 2030.
A recent history of troubling talking toys
The urgency of those rules is easier to understand after examining recent tests of children's AI devices. Testing published last December found serious problems in popular AI toys. One AI teddy bear gave detailed instructions on lighting a match. Two other toys discussed sexually explicit topics with researchers posing as users. The tests were designed to see whether safety filters could handle realistic conversations, and they showed that basic protections were not always working.
Just as concerning, the guardrails degraded over longer conversations. A toy might start with appropriate responses and then drift into unsafe territory after enough exchanges. This pattern is familiar in AI safety: short prompts may be filtered, but multi-turn dialogue can wear down the boundaries. The researchers who ran the tests found that four of the five toys appeared to use OpenAI models. OpenAI says its products are not intended for users under the age of 13, yet the toys were marketed to younger children.
The manufacturer of the teddy bear suspended sales, conducted an audit and later put the product back on the market. That response shows how improvised the current oversight can be. There is still no established playbook for recalling a toy because of the words it generates. Traditional product recalls focus on physical defects; conversational defects are harder to define, reproduce and fix.
Even social media companies have struggled to police AI interactions. Meta, for instance, has used contractors posing as teenagers to test rival chatbots. Those tests focus on safety, toxicity and age-appropriate behaviour, but they are expensive, uneven and difficult to scale. A toy that works differently every time a child uses it is a moving target for auditors.
A compliance strategy based on restraint
Luka's projector sidesteps a large part of this problem by not holding a conversation. The child speaks once and then draws. There is no assistant that can be persuaded to share instructions or engage in an inappropriate exchange. The AI has a narrow job: understand the child's prompt and identify a suitable outline. After that, the hardware is just a light source for a drawing activity.
That kind of bounded AI design may become a compliance advantage. As regulators ask difficult questions about data retention, mental health impact and model safety, manufacturers with limited conversational surfaces will have fewer risk points to defend. A drawing projector does not need to decide how to respond to a child's personal question or how to handle a sensitive topic. It does not build a relationship with the child, and it does not continue to operate while the child is asleep.
Regulators outside Europe are also taking notice of conversational AI. Canada has already extended social media age restrictions to AI chatbots, reflecting the view that chatbots can influence young users in ways similar to social platforms. If more countries follow, products that simulate friendship or offer open-ended chat will face additional gatekeeping. Luka's projector belongs to a category of AI toys that may be easier to place in children's hands because it does not attempt to become an autonomous companion.
The drawing projector is not a complete answer to every concern about AI and children. It still relies on speech recognition, model interpretation and projected output, all of which must be age appropriate. It asks for no ongoing conversation, but it still processes a child's voice, so privacy obligations remain. Parents may still want to know whether the prompt is stored, where it is stored and how long it is kept.
Those questions are now part of the mainstream conversation about connected toys, thanks in part to the EU's shifting regulatory landscape. Luka's projector shows one possible response: use AI to inspire, then step back. The model supplies a beginning, but the child controls the outcome. That may be good for creativity today and good for compliance tomorrow.