Microsoft Paint's new AI image generator builds on your brushstrokes

Feature & Demo

• Paint’s new “cocreator” mode turns rough brushstrokes into detailed images, similar to img2img/sketch workflows in Stable Diffusion.
• Several commenters found the official demo too short and unclear, making it hard to evaluate how much is user vs AI and how the workflow really feels.
• Some see it as genuinely useful for non-artists or mouse-only users who can’t draw precisely but want custom images beyond text prompts.

Local vs Cloud & Safety Checks

• One claim (citing an external writeup) says generation runs locally but each result is validated in the cloud for “safety,” causing latency.
• Another commenter asserts moderation itself is a local model on the NPU and that images are not uploaded; limited to NPU devices.
• This is unresolved in the thread; whether content is sent to Microsoft for checking remains unclear.

Business Model & Data

• Some argue cost is low because inference runs locally; the “payment” is buying new AI-PC hardware and staying on Windows.
• Others suspect broader motives: data collection, model training, lock-in, and long‑term monetization via subscriptions (e.g., O365/Windows-as-a-service).
• There is debate whether cloud AI services are currently run at a loss or already profitable, with no consensus.

Artistic Impact & Skill Development

• Enthusiasts: lowers the barrier to visual creation, helpful for engineers/game designers or kids who want good-looking art without years of training.
• Skeptics: liken it to paint-by-numbers or tracing; worry kids may never develop drawing or aesthetic judgment and that output will converge to an “AI look.”
• Some argue tools have always shifted required skills (calculators, Photoshop); creativity and taste will remain the real differentiators.

User Control, Censorship & Privacy

• Strong backlash against remote “safety” checks: concerns about censorship, puritanical filters, and precedent for Word/Notepad-style content policing.
• Comparisons to dumb tools (paint, cameras, drills) vs generative models that “create from thin air,” with associated legal and ethical liability.
• Worry that moderation may involve human reviewers or low-wage workforces viewing private images.

Alternative Tools & Hardware Constraints

• Multiple recommendations for Krita with Stable Diffusion / ControlNet as a local, FLOSS alternative that doesn’t “phone home.”
• Running quality diffusion locally currently favors PCs with dedicated GPUs; integrated graphics and most laptops struggle.
• Discussion of upcoming NPUs and unified-memory architectures (e.g., ARM/Apple-like designs) potentially making on-device generative AI more mainstream.

AI Hype & Product Direction

• Many see this as “AI everywhere” checkboxing and bloat in a tool valued for its simplicity.
• Others like Paint as the right place for playful AI — fun for kids and low-stakes experimentation.