NVIDIA Doosan physical AI tie-up and RTX Spark debut, Gemma-4 trends, Replicate drops

What shipped

On 7 June NVIDIA detailed fresh hardware and partnership moves while smaller releases appeared on Hugging Face and Replicate. The day mixed large-scale factory AI plans with quick-to-try models for agents and image work. Builders now have both enterprise-scale signals and immediate playground options to test.

Vendor launches

NVIDIA drove the section with two separate announcements. One links its full-stack platforms to Doosan robotics and factory projects. The second brings RTX Spark hardware to Korean gaming venues through local developers.

• •NVIDIA Doosan Partnership NVIDIA teamed with Doosan Group to combine its accelerated computing with Doosan robotics and power systems for factory automation projects, giving builders a concrete route to test physical AI on industrial equipment.
• •RTX Spark Rollout NVIDIA introduced RTX Spark to South Korean PC bangs via KRAFTON, NC and T1, letting creators run personal AI agents on gaming PCs without waiting for new hardware cycles.

Hugging Face trending

Gemma-4 Quantized Model: unsloth released gemma-4-12B-it-qat-GGUF on Hugging Face, a 12B model that supports fine-tuning and inference on modest hardware and gives vibe builders an open alternative to closed 10-15B options.

Replicate new models

Two new endpoints appeared on Replicate. One targets character generation while the other adds depth estimation for images.

• •Character Hater Model dannyboy2,323 shipped character-hater on Replicate, an endpoint that accepts prompts and LoRA weights so non-coders can generate or tweak character outputs through a simple web or API call.
• •Depth Pro Model oitoito added depth-pro on Replicate, an image-to-depth tool that processes a single photo and returns depth maps developers can drop straight into 3D or AR prototypes.

What this means for you

For Vibe Builders: You can now test physical AI ideas through NVIDIA and Doosan factory examples while running the new Gemma-4 checkpoint locally and calling the Replicate character and depth endpoints without writing servers. These give direct ways to add agent behavior or vision steps to side projects this week.

For Non-techies: For day-to-day work the new Replicate tools let you generate character images or depth maps with a few clicks, and the Gemma-4 model can run on a regular laptop for quick text or mixed tasks. NVIDIA news shows larger companies moving AI into real equipment, a signal that similar tools will reach smaller businesses soon.

For Developers: Production stacks can now pull the quantized Gemma-4 model for local agents and route vision calls to the depth-pro endpoint on Replicate while watching how NVIDIA RTX Spark hardware performs in gaming venues. Benchmark the 12B model against your current inference setup before swapping it into live pipelines.

What to watch next

Track follow-up NVIDIA factory deployments and any new quantized releases on Hugging Face. Check whether the Replicate endpoints gain usage volume or LoRA examples in the next few days.

Sources

Vendor launches

• •NVIDIA and Doosan Group Collaborate to Advance Physical AI and AI Factory Infrastructure
• •NVIDIA, KRAFTON, NC and Reigning ‘League of Legends’ Champions T1 Celebrate RTX Spark at Korea’s PC Bangs

Hugging Face trending

• •gemma-4-12B-it-qat-GGUF by unsloth trends on HuggingFace

Replicate new models

• •character-hater by dannyboy2,323 launches on Replicate
• •depth-pro by oitoito launches on Replicate