Local Image and Video Generation on AMD Ryzen™ AI Max+ Processor (Windows)

2026-07-14 · Source: AMD ROCm Blogs · Field: Technology & Digital — Artificial Intelligence & Machine Learning, AI Hardware Acceleration · Depth: Intermediate, long

Summary

AMD has enabled native Windows generative AI image and video workflows on its Ryzen™ AI Max+ processors, eliminating the need for Windows Subsystem for Linux (WSL). This is achieved through AMD ROCm™ 7.2.1 and a matching PyTorch build, which now run directly on Windows, driving the integrated AMD Radeon™ 8060S GPU. The key advantage is the processor's unified memory architecture (UMA), allowing the integrated GPU to access a large memory pool, typically around 94 GB. This capacity enables running memory-intensive models like full-precision Flux.1-dev (peaking ~34 GB) and WAN 2.2 text-to-video (peaking ~42 GB) without quantization or offloading, which would not fit on typical discrete GPUs with 8-24 GB VRAM. Benchmarks show SDXL generating at 1.48 it/s (18.41s end-to-end), Flux.1-dev at 3.64 s/it (77.56s end-to-end), and WAN 2.2 video in 26 minutes 51 seconds for 81 frames.

Key takeaway

For AI Engineers or Creative Technologists building local generative AI workflows on Windows, the AMD Ryzen™ AI Max+ processor with ROCm™ 7.2.1 offers a compelling native solution. You can now run memory-intensive models like full-precision Flux.1-dev and WAN 2.2 video directly, bypassing WSL and typical VRAM limitations. This simplifies setup and enables complex local creative work on a single, quiet system. Consider configuring your BIOS for maximum GPU-addressable unified memory to support larger models.

Key insights

AMD Ryzen™ AI Max+ processors enable native Windows generative AI workflows with large unified memory, bypassing WSL and VRAM limits.

Principles

• Unified memory expands GPU capacity.
• Native ROCm simplifies Windows setup.
• Integrated GPUs offer practical local AI.

Method

Install Python 3.12, create a virtual environment, then `pip install` AMD ROCm™ 7.2.1 and PyTorch 2.9.1+rocm7.2.1 from AMD's repo, followed by ComfyUI with a constraints file.

In practice

• Run SDXL for fast image iteration.
• Use Flux.1-dev for high-quality stills.
• Generate short-form video with WAN 2.2.

Topics

• AMD Ryzen AI Max+
• ComfyUI
• ROCm 7.2.1
• Unified Memory Architecture
• Local AI Inference
• Windows Generative AI

Code references

• comfyanonymous/ComfyUI
• comfyanonymous/ComfyUI

Best for: AI Engineer, Machine Learning Engineer, Creative Technologist

Related on AIssential

• AI Inference on AMD Ryzen™ AI Max Processor — AMD Ryzen™ AI Max+ processors enable local 100B+ parameter LLM inference via Unified Memory Architecture, reducing cloud dependency.
• Accelerating ComfyUI Workflows on AMD Instinct™ MI355X GPUs with ROCm — AMD Instinct MI355X GPUs significantly accelerate ComfyUI generative AI workflows over NVIDIA B200.
• Your AI Setup Needs This — AMD Ryzen AI Processors integrate NPU, GPU, and CPU for enhanced on-device AI performance in business laptops.
• AMD unveils $3,999 Ryzen AI Halo PC for local developer workloads — AMD's Ryzen AI Halo PC offers a cost-effective, local AI processing alternative to cloud services for professionals.
• Warning Shots Fired as AMD Announces New Data Center GPUs — AMD is aggressively challenging Nvidia's AI hardware dominance with new GPUs, CPUs, and an open software ecosystem.
• AMD's tiny AI PC points to a more local future for model inference — Local AI inference on compact hardware reduces cloud dependency, improves privacy, and lowers latency.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AMD ROCm Blogs.