It's a LoRA with 13 school square pools studied.The dataset did not converge nicely unless the shape and composition of the data sets were consistent! My article explains this.Features・Potentially perspective background.・You can create a poolside roof.<hr />Strength:0.8Recommended prompts⬇<pre><code>outdoors,pool ladder, poolside, white pool ladder,reflected water,blue sky,white roof,tree</code></pre>My reserch (Low LR ver)<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/08398b91-ce72-4f40-a0cc-1db9bf5aad26/width=525/08398b91-ce72-4f40-a0cc-1db9bf5aad26.jpeg" />Comparison.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/38118cf5-8391-4bdb-a6cb-61f4286b52cd/width=525/38118cf5-8391-4bdb-a6cb-61f4286b52cd.jpeg" />It can be seen that changing the LR makes almost no difference.The change is that learning slows down and takes more steps to complete.back number: <a target="_blank" rel="ugc" href="https://civarchive.com/models/1600795/poolfor-research">https://civarchive.com/models/1600795/poolfor-research</a>The dataset did not converge nicely unless the shape and composition of the data sets were consistent! My article explains this.<hr />The thumbnail video uses a frame pack specifically for background rotation.I also explain this in my article ⬇<a target="_blank" rel="ugc" href="https://civarchive.com/articles/15090/framepacklora">https://civarchive.com/articles/15090/framepacklora</a>Note⬇<a target="_blank" rel="ugc" href="https://note.com/tori29umai/n/ncf87a3355ac7">https://note.com/tori29umai/n/ncf87a3355ac7</a><a target="_blank" rel="ugc" href="https://huggingface.co/spaces/tori29umai/FramePack_rotate_landscape">https://huggingface.co/spaces/tori29umai/FramePack_rotate_landscape</a>

School Pool V2 (学校のプール)

13images, 4repeat, 34epoch,3D model and illustration, with the addition of humans.I Show Sample of Dataset in the explanation.

Files with SHA256: ed76fdfd6d810cffc64d84fe34ce5d2f504617b1d2a75e15f8f06423e9bc5c69