Open GaidamakUA opened 8 months ago
I fixed it by downloading the fixed vae model following this:
SD1.x, SD2.x and SDXL support
!!!The VAE in SDXL encounters NaN issues under FP16, but unfortunately, the ggml_conv_2d only operates under FP16. Hence, a parameter is needed to specify the VAE that has fixed the FP16 NaN issue. You can find it here: SDXL VAE FP16 Fix.
you may try. However, I got a very strange colored image with that vae.
This command is working fine, even though it's kind of slow 16 s/it
./bin/sd -M txt2img -m ../models/checkpoints/reliberate_v20.safetensors -p "Two girls playing tennis"
This command produces black images, but iterates faster 8 s/it./bin/sd -M txt2img -m ../models/checkpoints/reliberate_v20.safetensors -p "Two girls playing tennis<lora:lcm_lora_1.5:1>" --lora-model-dir ../models/loras --steps 4 -v --cfg-scale 1 --sampling-method lcm
I don't know if it's relevant but it gives me this output