<h2 id="ga1n3rb0t_v1.02-hunyuan-video-hm8wqrv4g">ga1n3rb0t_v1.02 Hunyuan Video</h2>ga1n3rb0t v1.02 is a Lora I created for the Hunyuan t2v Video Model, trained using the diffusion-pipeline. This Lora was trained with a combination of short video clips and images utilizing an H100 80gb card.Use "ga1n3rb0t" for the trigger in prompts. General: ga1n3rb0t v1.02 specializes in producing semi-realistic, heavy-set, thick, curvy women, ranging from plump to very fat. This Lora was trained on videos of women interacting with their belly - grabbing, rubbing, squeezing - a concept it seems to have learned very well. It was also trained on videos of weight gain and expansion - though it has not been as successful outputting results for this category (It can do it, but it is difficult to trigger for some reason). The rest of the dataset was comprised of 1024x1024 and 768x1024 images output from a Pony Lora I developed.Prompting Tips: <ul><li>Use "plump" and "chubby" for smaller. Use "fat" and "huge belly" to get larger.</li><li>Lora weight of 1.00 works great, but lowering the weight to 0.7-0.8 can yield more realism if that's what you're looking for.</li><li>If you are wanting a fully-clothed result (i.e. belly not exposed), note it near the beginning of your prompt for a higher rate of success, as the model is biased towards skimpier attire. If you are wanting skimpy attire (lingerie/bikini/naked) you can throw it in pretty much anywhere and the model will pick it up.</li><li>Interaction with belly and breasts has been very easy to achieve, simply prompting using words like "grabbing", "squeezing", "rubbing" belly/breasts.</li><li>If you are interested in trying to achieve weight gain/expansion output, the prompts for that training data included the phrases: "rapid weight gain", "belly expansion", and "breast expansion". I have found a success rate of around 1/10 generations so far, but perhaps you will have more success with better prompting. </li></ul>Thanks for testing it out, I hope you enjoy! Looking forward to seeing what you can achieve with this Lora. Feedback is welcome!

ga1n3rb0t v1.02 for Hunyuan Video

Version 1.02 is the first to be made public. Version 1.01 (never published) achieved limited success during testing due to too many still/stagnant video frames in training data - this resulted in often low movement during testing. For v1.02 training videos were trimmed to be much shorter (25-40 frames) and any still/stagnant frames that did not translate movement from one frame to the next were removed.

Files with SHA256: 8847a922ff40fd07ccc80913627ff37b3dbb4d2cd63e831ca52f357e861ac314