The default FLUX DEV model lacks diversity in many aspects in the moment. Therefore, using cheap frontal flash photography as training data for a LoRa model seems an effective way to explore the bandwidth of the FLUX DEV model. This LoRa training experiment aims to challenge both the stereotypical visuals of FLUX and the representation of age and gender in its motifs. Consequently, the training data was specifically selected to include images of diverse objects, scenes, and people, captured using blurry, grainy, cheap flash-based analog photography. Special attention was given to images where isolated objects are highlighted by the flash while the rest of the image fades into darkness due to the lack of light. Mostly frontal flashed suburban and wildly overgrown forest scenes are in the training data + cropped wild perspectives of people.
Description
first shot of training with Replicate LoRa trainer
FAQ
Comments (2)
can you provide text prompts? I am having trouble duplicating your photo style
Sure that. The LoRa is my first shot with FLUX. The trigger word in front of prompt seems to have a good effect in my tests. An example prompt is: " fluxflashy analog grainy photo of a casual short haired old bored female person in suburban concrete wall with weird crappy paintings overgrown with wilderness and distressed and dirty, barefeet, realistic skin, casual clothes from the 80s east europe, many colorful lilies blossoms covering all the face of the female person who is in dynamic weird poses from the side or from the back, person bend over to the front, there is motion blur, it is allmost dark and after sunset, image is out of focus, vignette, depth of field, white smoke or fog on the ground, dirt on lens "
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.