This is basically a stabilization of <a target="_blank" rel="ugc" href="https://civarchive.com/models/833294?modelVersionId=962003">the v-prediction test of NoobAI</a>. Weights from <a target="_blank" rel="ugc" href="https://huggingface.co/comin/IterComp">IterComp XL</a> were added onto the model using this method, then the original clip was added back in.Perpendicular component is used with add difference to prevent interference with v-prediction noise.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/b0606d3b-7172-4e69-a01d-e7f2cc2a48bc/width=525/b0606d3b-7172-4e69-a01d-e7f2cc2a48bc.jpeg" />Nodes can be found here (requires <a target="_blank" rel="ugc" href="https://github.com/ljleb/sd-mecha">sd_mecha</a> installed): <a target="_blank" rel="ugc" href="https://github.com/ljleb/comfy-mecha">https://github.com/ljleb/comfy-mecha</a>Iterative Composition-Aware Feedback Learning: <a target="_blank" rel="ugc" href="https://arxiv.org/pdf/2410.07171">https://arxiv.org/pdf/2410.07171</a>V-prediction support has been added to Automatic1111 recently. On ComfyUI, use the ModelSamplingDiscrete node.

NoobAI VPred IterComp Test

<a target="_blank" rel="ugc" href="https://civitai.com/models/833294?modelVersionId=962003">v-pred test version</a>

Files with SHA256: e258db2938efa9ca391993a431df49a44bba9bfa9d31069a6610530af1287923