Workflow: ClownsharKSampler -> Upscale Latent By 2 -> ClownsharKSampler. Prompt more or less with LM-Studio - qwen3-vl-8b-nsfw-caption-v4.5. So, image after image, NO art whatsoever. On Runpod with 4000 - 500s, 4090 - 165s, 5090 - 120s, 6000 - 80s. And maybe some day I find out cause of Qwen lines (incompatible Loras).