Yes, I am doing that now on most smaller resolution datasets for concepts, however I would be very w
Yes, I am doing that now on most smaller resolution datasets for concepts, however I would be very wary of any noticeable artificial enhancements it may be doing to your dataset as that "artificialness" may be enhanced in your SD outputs when you include the LoRA in your inference. I find that keeping the prompt text influence fairly low ~4 or less seems to keep the upscale/detail recovery adequate without introducing artificialness into the AI upscaled image. Your mileage may vary...
