>>26420
By limiting the length to 15 you’ll be able to bump the resolution by 0.1 or even 0.2 mp since the 0.3 you are currently using is extremely low compared to the model’s training resolution. Also use kijai nodes related minimax v-ram nodes to better handle memory and update comfy to latest version and start it up using that incredible comfy-kitchen attention they’re just developing which is faster than sage attention while being also higher quality