Int8/int6 mixed ckpt, with turbo LoRA from PrunaAI. Fits 8GiB GPU without swapping.Uploaded here for convenience. Avoid additional quantization loss when applying the LoRA on quantized ckpt (weights got quantized twice).The turbo LoRA was merged with 0.75 strength. Thus needs 16 steps. This is the sweet spot I found. Stabilized image a lot, while no editing downgrade.All credit belongs to the original model author. License is the same as the original model.<ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/Qwen-Image-2.1/blob/main/diffusion_models/qwen_image_2.1_bf16.safetensors">https://huggingface.co/Comfy-Org/Qwen-Image-2.1/blob/main/diffusion_models/qwen_image_2.1_bf16.safetensors</a></li><li><a target="_blank" rel="ugc" href="https://huggingface.co/PrunaAI/Pruna-Qwen-Image-2.1/blob/main/p_qwen_image_2.1_8step_v0.1.safetensors">https://huggingface.co/PrunaAI/Pruna-Qwen-Image-2.1/blob/main/p_qwen_image_2.1_8step_v0.1.safetensors</a></li></ul>

Files with SHA256: 65067368a61bd33e6fbccf739db59dd91a03e37ae81a9e14b64f48c7fd3c8eac