<h2 id="unipixel-style-experiment">UniPixel Style 🎨 [Experiment]</h2>--[<a target="_blank" rel="ugc" href="https://tensor.art/models/919932018832188492">Usable on TensorArt</a>]--this a sub experiment of my other <a target="_blank" rel="ugc" href="https://civarchive.com/models/2024385?modelVersionId=2291120">UniPixel LoRA</a> trained on Wan2.2 T2V model, using a subset of image from the original treated dataset and some added pixelart videos edited to fit 16fps.<h3 id="usage-tips-(mini-ver)-:">Usage Tips (mini ver) :</h3><ul><li>something to note about Wan2.2 gens, the high noise pass adds motion but will a lot of noise the Low noise pass refine the result focusing on treating the details so:<ul><li>use both the High and Low for video generation (Low:1.0 , High:0.8)</li><li>use the Low only version for image generation</li></ul></li></ul><h2 id="observations:">Observations:</h2><ul><li>most gens looks too vibrant like like they're out of retro kids game</li><li>it doesn't handle fast movement well </li></ul><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/073c45f1-4b70-46ea-a6e2-c39b71e467a1/width=525/073c45f1-4b70-46ea-a6e2-c39b71e467a1.jpeg" />

UniPixel - Style

wan_t2v_h_unipixel-e6.safetensors

T2V High

T2V Low

Unipixel WAN 2.2 Experiment :Meant as final layer of processing (on the low noise model pass) that attempts to apply a pixel art like style on the animation/image. tho it might be a bit of a stretch I'll see if can capture the choppy pixel motion in an other LoRA mean for the high noise pass .

Files with SHA256: 9b30492e8f7921ba274310ce3e9f0ad281647df9cc48040e9b54bc8209f5a1c0