This is a LoRa model for generating film-style AI photos. It has been downloaded nearly 60,000 times on the C site, and has been well received by everyone.

This model can be used with realistic models such as MoonMix, MajicMix Realistic, or ChillouMix to produce realistic film-style photos. It is also recommended to use HIGHLY INVERSION models: bad_pictures, negative_hand negative embedding. These two negative embedding

models do not significantly affect the texture of the film.

2023.7.8 Update “The VELVIA 3.0”: The number of images in the training set for Velvia 3.0

's main LoRa model has reached 406.

• The resolution was further increased to 1024*1024
.
• The main model training in this version uses the scale weight norm function to avoid overfitting the model. I also tested the “Prodigy” optimizer in another backup version. This optimizer can provide a level of fit comparable to AdamW8bit at half of the total number of training steps, but maybe my parameter settings were inappropriate and the training results had
poor facial beauty, so I didn't use it in the end.

Based on the high-resolution subject model, the Velvia 2.0 model was incorporated in a small proportion, while layered adjustments were made to improve the portrait's skin tone.

Update Content:

• The number of training images for the main LoRa model of Velvia 3.0 has improved 406, with the resolution Further explanation to 1024x1024
.
• In this version, the main model training utilizes “Scale Weight Loss” to avoid overfitting. I also tested the “Prodigy” Optimizer in another alternative version. Prodigy can provide a fitting degree comparable to ADAMW8bit with half the total training steps, but due to precise parameter settings, the facial aesthetics of the training The results were not as good, so it was not surprising in the end.
• Based on the high-resolution main model, a small variation of the Velvia 2.0 model was blended, and hierarchical considerations were made to improve the skin Tone in portraits

.