So SD3.5 not worth it with less than 32gb of vram?
anyone try this:
Low RAM Solution
If the generation crashes, it likely indicates insufficient RAM.
Use fp8_scaled workflow json (experimental) and the fp8 scaled model as a low vram option
You can try t5xxl_fp8_e4m3fn_scaled.safetensors or t5xxl_fp8_e4m3fn.safetensors instead of t5xxl_fp16 for lower memory usage (the scaled one is an experimental checkpoint by us). But t5xxl_fp16 is still recommended if you have more than 32GB RAM.