fp8 and gguf workflow (single image)

• Qwen2509_fp8_16GB_12GB_GGUF_6GB_4steps_v8 w/Lightning Lora

• Qwen2509_fp8_16GB_12GB_GGUF_6GB_20steps_v8

VRAM usage (second pass)

• 16GB : bf16 (380 seconds on 4070tis @50steps 24GB MEM ,36GB MEM@VAE)

• 16GB : fp8 (25 seconds on 4070tis @4steps)

• 12GB : fp8 w/DisTorch2MultiGPU

• 6GB : Q4gguf w/DisTorch2MultiGPU

• 16GB : Q4gguf

• 16GB : fp8 meitu (160 seconds on 4070tis @20steps) (I felt the prompt response was 2509 >> Meitu)

I'm uploading this because it was hard to find a workflow that doesn't use Lightning Lora.

(But Lightning 4steps is very good.)