AMD Announces Amuse 3.0 Generative AI Solution for Print-Quality Images and Draft-Quality Short Videos

[btarunr]

AMD partnered with TensorStack AI to launch the Amuse 3.0 local generative AI platform, letting users of AMD Ryzen AI processors and Radeon RX GPUs generate images and short-videos locally. The platform can generate print quality images and draft-quality (read: low res) videos up to 6 seconds in length. The platform comes with support for over 100 new AI models, including Stable Diffusion 3.5 and FLUX.

Each of these models has been heavily optimized for AMD hardware resulting in up to 4.3 times faster inferencing speed compared to generic models. The platform can diffuse 4-6 seconds of draft-quality short videos. It also includes AI-based video filters. AMD arrived at the 4.3x performance number comparing generic Olive Optimize base models with AMD optimized models, accelerating both an AMD Radeon RX 9070 XT GPU. The company also put out performance numbers for Ryzen AI processors with their 50 TOPS-class NPU, where the AMD Optimized models were shown being up to 3.3x faster with image generation.



The complete slide deck follows.



View at TechPowerUp Main Site

 

[Steevo]

They are so close to making it relevant for more than Christmas cards. If they could make a one click installer than ran and could create short generated descriptive videos "storyboard" for movie making, and/or if they made longer videos possible by feeding subject matter into a truly local model for small creators and possibly the ability to create unique blockchain tokens for content created that would be the nugget they need to make this truly relevant. As of now its just a toy it feels like.

 

[Lesha]

3.0 is not available yet on their website.

 

[john_]

They are so close to making it relevant for more than Christmas cards. If they could make a one click installer than ran and could create short generated descriptive videos "storyboard" for movie making, and/or if they made longer videos possible by feeding subject matter into a truly local model for small creators and possibly the ability to create unique blockchain tokens for content created that would be the nugget they need to make this truly relevant. As of now its just a toy if feels like.

Amuse IS one click installer.

 

[Makaveli]

3.0 is not available yet on their website.

This.

 

[ZoneDymo]

just pointless, its as if people fill their lives with MS paint.....

 

[Makaveli]

just pointless, its as if people fill their lives with MS paint.....

Pointless based on what?

 

[Visible Noise]

And yet RoCM support is nowhere to found for current generation AMD video cards.

What a farce.

 

[Firedrops]

It's hilarious how much effort AMD is willing to pour into niche gimmick features as opposed to properly supporting their whole rocm, torch-directml ecosystems.

This program is going to be 100% useless if it doesn't also have near-universal support for things like community fine-tuned models, loras, controlnet. It sounds like it only supports the base models that AMD has converted with their internal secret techniques, and AMD AI announcements for the past 10 years have been overpromise and underdeliver.

 

[Visible Noise]

Bwaaahahahahaa… RX 9000 isn’t recommended.

God AMD continues to blow their own brains out.

 

[umeng2002]

We truly are in the AI slop era.

 

[igormp]

3.0 is not available yet on their website.

They finally updated the website with the latest version.

However, 9000 series is still not in the recommended specs list.

 

[Bruno Vieira]

They finally updated the website with the latest version.

However, 9000 series is still not in the recommended specs list.

It requires too much VRAM; the minimum recommended spec is 24 GB, and you will often exceed it.

 

[igormp]

It requires too much VRAM; the minimum recommended spec is 24 GB, and you will often exceed it.

It depends on which kind of model you're trying to run. The smaller, quantized models should require way less than that.