Blog
Written versions of our YouTube videos, so you can check the numbers and tables at your own pace.
-
Mac mini M6 AI Image: 3 min 20 s to 2 min 20 s, Qwen-Image 2.1 Measured (youtube.com)
On a Mac mini M6 32GB, Qwen-Image 2.1 took 3 min 20 s in default ComfyUI, about 2 min 20 s with just the model file switched to bf16, and about 2 min with mflux 4-bit. Korean text accuracy and memory picks included.
-
MiniMax H3 Best Settings on One RTX 5090: 7 Problems Making an AI Short (youtube.com)
The MiniMax H3 settings chosen while making a 20-minute AI short on one RTX 5090, fixes for repeated dialogue, stray voices and wide-shot positions, and measured SageAttention results.
-
Free AI Video Upscaling Measured: FlashVSR vs SeedVR2, 25 Minutes and 3 Minutes Looked Identical (youtube.com)
We timed the free upscalers FlashVSR and SeedVR2 on an 8GB laptop and an RTX 5090. tiny-long matched tiny pixel for pixel yet ran 8.7x faster, and settings made no difference.