{"ModelId":766608,"ModelName":"SameFace Fix [Flux Lora]","ModelDescription":"\u003Ch3 id=\u0022important:-this-lora-should-be-used-with-a-negative-weight.-6sf4ejr5z\u0022\u003EImportant: \u003Cspan style=\u0022color:#fa5252\u0022\u003EThis Lora should be used with a \u003C/span\u003E\u003Cstrong\u003E\u003Cem\u003E\u003Cspan style=\u0022color:#fa5252\u0022\u003ENEGATIVE\u003C/span\u003E\u003C/em\u003E\u003C/strong\u003E\u003Cspan style=\u0022color:#fa5252\u0022\u003E weight.\u003C/span\u003E\u003C/h3\u003E\u003Cp\u003EAlso: The model is aimed at diversifying the look of women only. I plan to do a separate model for men later (and men with beards especially)\u003C/p\u003E\u003Cp\u003EAlso cool fact: the size of the Lora is only 4mb (it is limited to 2 blocks with a high rank)\u003C/p\u003E\u003Cp\u003E\u003C/p\u003E\u003Cul\u003E\u003Cli\u003E\u003Cp\u003ESo, What\u0027s a \u0022Sameface\u0022?\u003C/p\u003E\u003C/li\u003E\u003C/ul\u003E\u003Cp\u003EIt\u0027s a common problem in generative AI, in which the model generates people with identical appearances. Unlike with SD1.5/SDXL, this is not just a result of the diverse dataset converging into a single token like \u0022woman\u0022. in Flux, Dev/Schnell were distilled from the Pro version, resulting in a loss of diversity.\u003C/p\u003E\u003Cp\u003E\u003C/p\u003E\u003Cul\u003E\u003Cli\u003E\u003Cp\u003EWhy use Negative weight?\u003C/p\u003E\u003C/li\u003E\u003C/ul\u003E\u003Cp\u003ELora was trained on the most generic photos of women that Flux generates.\u003C/p\u003E\u003Cp\u003EIf you then apply this Lora with a NEGATIVE weight, it will work as a sort of negative prompt, but without negative prompt!\u003C/p\u003E\u003Cp\u003EBy using this clever trick, Lora will basically tell Flux to generate anything except a typical Flux face. And as result, we\u0027ll get beautiful unique faces every time!\u003C/p\u003E\u003Cp\u003E\u003C/p\u003E\u003Cul\u003E\u003Cli\u003E\u003Cp\u003EWhy Lora and not tricks like lower Guidance?\u003C/p\u003E\u003C/li\u003E\u003C/ul\u003E\u003Cp\u003EPeople want diverse faces with high guidance in order to maintain high prompt adherence and image quality, that\u0027s why interventions at the latent space level are optimal.\u003C/p\u003E\u003Cp\u003E\u003C/p\u003E\u003Cul\u003E\u003Cli\u003E\u003Cp\u003EHow does it differ from other character Loras?\u003C/p\u003E\u003C/li\u003E\u003C/ul\u003E\u003Cp\u003EWhile other Loras introduce their own \u0022samefaces\u0022, this Lora, only tells Flux what should NOT be in the generation, not what to add. As a result, this Lora diversifies people\u0027s appearance, rather than narrowing it even further.\u003C/p\u003E\u003Cp\u003E\u003C/p\u003E\u003Cul\u003E\u003Cli\u003E\u003Cp\u003EHow was the Lora actually made?\u003C/p\u003E\u003C/li\u003E\u003C/ul\u003E\u003Cp\u003EAt first I was disappointed because nothing worked out. I was about to give up this project, but then tried to train Lora with just 2 layers, and suddenly during testing, I was met with beautiful faces, and with authentic, distant from Flux, appearance.\u003C/p\u003E\u003Cp\u003E\u003C/p\u003E\u003Cul\u003E\u003Cli\u003E\u003Cp\u003EWhat\u0027s next?\u003C/p\u003E\u003C/li\u003E\u003C/ul\u003E\u003Cp\u003EI have a lot to experiment with. Different learning rates, layers and ranks have to be tested in order to improve quality. Merging different Lora variants might fix the problems with artifacts at higher values. I\u0027ll make sure to make a better version, If this Lora has enough interest\u003C/p\u003E\u003Cp\u003E\u003C/p\u003E\u003Cp\u003E\u003Cstrong\u003EI\u0027m really forward to your feedback, as this model is highly experimental. Any feedback is appreciated and will hopefully help me to improve the model.\u003C/strong\u003E\u003C/p\u003E\u003Cp\u003E\u003C/p\u003E\u003Cp\u003E\u003Cem\u003Especial thanks to Ostris for \u003C/em\u003E\u003Ca target=\u0022_blank\u0022 rel=\u0022ugc\u0022 href=\u0022https://github.com/ostris/ai-toolkit\u0022\u003E\u003Cem\u003EAI-Toolkit\u003C/em\u003E\u003C/a\u003E\u003C/p\u003E","Nsfw":false,"Tags":["character","woman"],"ModelType":"LORA","VersionId":857446,"VersionName":"v0.1","VersionDescription":"\u003Cp\u003EFirst Version!\u003C/p\u003E","BaseModel":"Flux.1 D","FileMetadata":{"fp":null,"size":null,"format":"SafeTensor"},"ImportedAt":"2025-03-10T09:27:10.5277856+00:00","Hashes":{"SHA256":"5B7919B792DC536621D95D0F42F77D3D68789874F4D5926C42F7242C30F78E7C","CRC32":"7A6DCBD4","BLAKE3":"B772A4E718B9B861A879DF95BE6CE1EAE1AB20E953470253AE1BE6B135E6C337"},"TrainedWords":["woman"],"Stats":{"favoriteCount":0,"commentCount":18,"thumbsUpCount":693,"downloadCount":7866,"ratingCount":0,"rating":0},"UserTitle":null,"ThumbnailImageUrl":null}