>>106407231
>after years, proprietary services have better and better image generation
>literally write a shit prompt and it is 99% of the time nice
>with extreme amount of tinkering, 8x3090 can't generate anything right
>local models outputs blurry shit
>now those eight 3090s feels like wasted money and gemini-2.5-flash-image-preview was released
>now you can do even simpler shittier prompts and gemini-2.5-flash-image-preview just creates images that are 100% perfect every time
How to cope with this shit? Can I run something that is like gemini-2.5-flash-image-preview locally? All this shit is so bad. DeepSeek R1 is perfect but we have to have over 3 TB RAM to run it locally with full precision.