| {"ModelId":796382,"ModelName":"UltraRealistic Lora Project","ModelDescription":"\u003Cp\u003EThe main goal was to make more \u0027live\u0027 images with livelier emotions and more dynamic poses and slightly amateurish quality (also can produce high quality images). \u003Cbr /\u003E\u003Cbr /\u003EP.S: Thanks to everyone for the feedback! I\u0027ve noticed the comments (not only here) about anatomy issues, and I\u2019ve been collecting reports and examples from all of you. I\u0027m considering a couple of options to address this:\u003C/p\u003E\u003Col\u003E\u003Cli\u003E\u003Cp\u003ETraining a full checkpoint, which might be the more rational approach for consistent improvements.\u003C/p\u003E\u003C/li\u003E\u003Cli\u003E\u003Cp\u003EExpanding the dataset with more photos that cover diverse poses to help refine anatomical accuracy.\u003C/p\u003E\u003C/li\u003E\u003C/ol\u003E\u003Cp\u003EI appreciate the input - it really helps me shape the future updates\u003Cbr /\u003E\u003Cbr /\u003EP.S.2: I\u2019ve moved training to RunPod for consistent, high-quality results. If you\u0027d like to support my work and future updates, you can find me on Ko-fi. Cause i plan to fine-tune a model, not just a LoRa. \u003Ca target=\u0022_blank\u0022 rel=\u0022ugc\u0022 href=\u0022https://ko-fi.com/danrisi\u0022\u003Ehttps://ko-fi.com/danrisi\u003C/a\u003E\u003Cbr /\u003E\u003Cbr /\u003ESetting i use in ComfyUI for Flux:\u003Cbr /\u003ECFG=1, Guidance =2.5, Scheduler=Beta, Sampler=dpmpp_2m, Steps=40, Strength= from 0.8 to 1.0 works good, but i\u0027m usually use with 1, but if hands getting worse then i set 0.87\u003Cbr /\u003E\u003Cbr /\u003ESetting i use in ComfyUI for SD3.5:\u003Cbr /\u003ECFG=1, Guidance =3.5, Scheduler=sgm_uniform, Sampler=dpmpp_2m, Steps=40, Strength= from 0.5 to 1.0 works good, but i\u0027m usually use with 0.7\u003Cbr /\u003E\u003Cbr /\u003EV2 - Flux\u003Cbr /\u003E\u003Cbr /\u003EBrings even more realism and versatility to your creations, with significant improvements in stability, anatomy, and overall quality. This update makes the LoRA more adaptive, allowing you to achieve various quality levels based on your prompts\u2014from high-definition realism to intentionally lower-quality aesthetics.\u003C/p\u003E\u003Cp\u003ETrained on 1048 images.\u003C/p\u003E\u003Cp\u003E\u003Cstrong\u003EWhat\u0027s New:\u003C/strong\u003E\u003C/p\u003E\u003Cp\u003E\u003Cstrong\u003EStability Improvements\u003C/strong\u003E: The new version is more stable and works better with text-based prompts, providing a smoother and more predictable output.\u003C/p\u003E\u003Cp\u003E\u003Cstrong\u003EEnhanced Hands \u0026amp; Anatomy\u003C/strong\u003E: Hands and body anatomy are more refined, enhancing lifelike quality.\u003C/p\u003E\u003Cp\u003E\u003Cstrong\u003EQuality Flexibility\u003C/strong\u003E: With the right prompts, you can adjust for both high-quality and lower-quality aesthetics (examples available).\u003Cbr /\u003E\u003Cbr /\u003E\u003Cbr /\u003E\u003Cbr /\u003EV1.2 for SD3.5 - Large\u003Cbr /\u003E\u003Cbr /\u003EDecided to make a version for sd3.5 almost with same settings. Imo looks good, but i noticed some problems with anatomy (in some moments even worse then flux), but aesthetic (colors, contrast and other stuff looks even better then flux). I see that sd3.5 have potential, maybe new versions will be on sd3.5 too.\u003Cbr /\u003E\u003Cbr /\u003EV1.2\u003C/p\u003E\u003Cp\u003EChanged half of images in dataset, changed prompting style, improved hands, less \u0027flashlight effects\u0027 at night scenes and overall LoRa quality improvement (i hope). \u003Cbr /\u003EIn this version it\u0027s unnecessarily to use tons of \u0027trigger words\u0027 like in V1. Just add some of them that i mentioned in trigger words\u003C/p\u003E\u003Cp\u003EV1\u003C/p\u003E\u003Cp\u003ETrained another amateur lora (that\u0027s already been done a lot around here).\u003C/p\u003E\u003Cp\u003EStill, I took some pictures from my dataset on the 2000s and added another 700 pictures. It came out pretty good so far, but there is a controversial point about quality optimization, I was hoping it would help control the quality, but as I see it only confused the model. In the dataset there are many different gradations of quality, like:\u003C/p\u003E\u003Cp\u003E1) High-resolution photo, shot on a mobile phone, no visible artifacts, clear and sharp\u003C/p\u003E\u003Cp\u003E2) Low-resolution, amateur photo shot on digital camera, no visible jpeg artifacts, slightly noisy\u003C/p\u003E\u003Cp\u003E3) Medium-resolution photo, shot on a mobile phone, slight graininess due to low light conditions, no significant digital artifacts\u003C/p\u003E\u003Cp\u003EAnd other their combinations and variations. I think i\u0027ll remove such in next version. But this one working the best for me:\u003C/p\u003E\u003Cp\u003ELow-resolution, amateur photo shot on digital camera, no visible jpeg artifacts, slightly noisy\u003C/p\u003E","Nsfw":false,"Tags":["photorealistic","sexy","aesthetic","style","woman","girls","man"],"ModelType":"LORA","VersionId":1026423,"VersionName":"Flux - v2","VersionDescription":null,"BaseModel":"Flux.1 D","FileMetadata":{"fp":null,"size":null,"format":"SafeTensor"},"ImportedAt":"2025-03-13T07:36:40.7129999+00:00","Hashes":{"SHA256":"B1C4DDF95671E6B51817B4F3802865E544040C232C467E76B1CB0C251BD6B634","CRC32":"4D681E09","BLAKE3":"411049EDCEA4C65F520FEECF85D86F6450C303D580627FBA74F9F00198570531"},"TrainedWords":["amateurish photo","low lighting","in motion","overexposed","underexposed","GoPro lens","eerie atmosphere","smeared background","smeared foreground"],"Stats":{"favoriteCount":0,"commentCount":60,"thumbsUpCount":2275,"downloadCount":42938,"ratingCount":0,"rating":0},"UserTitle":null,"ThumbnailImageUrl":null} |