<h3 id="fp8-and-gguf-workflow-(single-image)">fp8 and gguf workflow (single image)</h3><ul><li>Qwen2509_fp8_16GB_12GB_GGUF_6GB_4steps_v8 w/Lightning Lora</li><li>Qwen2509_fp8_16GB_12GB_GGUF_6GB_20steps_v8</li></ul><h3 id="vram-usage-(second-pass)">VRAM usage (second pass)</h3><ul><li>16GB : bf16 (380 seconds on 4070tis @50steps 24GB MEM ,36GB MEM@VAE)</li><li>16GB : fp8 (25 seconds on 4070tis @4steps)</li><li>12GB : fp8 w/DisTorch2MultiGPU</li><li>6GB : Q4gguf w/DisTorch2MultiGPU</li><li>16GB : Q4gguf</li><li>16GB : fp8 meitu (160 seconds on 4070tis @20steps) (I felt the prompt response was 2509 &gt;&gt; Meitu) </li></ul>I'm uploading this because it was hard to find a workflow that doesn't use Lightning Lora.(But Lightning 4steps is very good.)

i2i 6GB 12GB 16GB bf16 fp8 gguf 4steps 20steps (Qwen2509 Workflow)

Qwen2509_fp8_16GB_12GB_GGUF_6GB_20steps_v8_workflow

Files with SHA256: 97e5a5fc140aab18b2de0de92dc13ef0d4075dce44c86bd8555bee8a3a96137d