I tried to get this result without any LoRA but it seems almost impossible, this helps a lot even if the results are not always crystalline and flawless... Better than nothing :D
More training images improved the results on furniture
twistcolu-12.safetensors