<h3 id="include-the-year-you-want-to-target-in-your-prompt-like-2000-2001-2002-2003-2004-or-2005-and-either-'tv'-or-'commercial'-to-ensure-that-the-model-gets-included-in-the-output.-rglszrhe8">Include the year you want to target in your prompt, like 2000, 2001, 2002, 2003, 2004 or 2005, and either 'tv' or 'commercial', to ensure that the model gets included in the output.</h3>Trained against frames from tv commercials recorded onto VHS from my youtube channel, @Dadsweathertapes. Frames were upscaled from SD footage and retained haloing, chroma issues, and other VHS artifacts. The VHS tapes were digitized in the last 2 years for the above youtube channel, using a standard 4-head consumer VCR and a Collossus II input card.Training data used for this particular LoRA were dated 1994 through 1999.Dataset preparation using random segments of video were prepared using scripts from my github - <a target="_blank" rel="ugc" href="https://github.com/akspa0/dataset-tools/blob/main/video/randomSampler/videosampler2x6.py">https://github.com/akspa0/dataset-tools/blob/main/video/randomSampler/videosampler2x6.py</a>

retroVHS:2000-2005

epoch 37; it was trained to 62 epochs but looks very over-trained and it loses coherency

Files with SHA256: b088d01a7d2292afca8ef602f7027aee9744227014a9dfae72190e99ff1f7d05