santa hat
deerdeer nosedeer glow
Sign In
Flux LoRA's for Realism

On my quest for realism, I tried mixing these two LoRAs:

Phlux - Photorealism with style v1

Amateur Photography [Flux Dev] v2

All images generated used these settings:

  • Flux Dev

  • t5xxl_fp8

  • Fixed Noise Seed

  • 896 x 1152

  • Sampler: dpmpp_2m

  • Scheduler: sgm_uniform

  • 20 Steps

  • Guidance: 3.5

(Guidance: 2.7 was giving bad results)

Prompt: An amateur photo of a 40-year-old Japanese business man taking a selfie in front of a window in an office in a skyscraper in Tokyo. He is wearing a navy blue suit. There is a glare and a lens flair from the early evening sun. It is partly cloudy.

Phlux Weight: 0.00

Amateur Weight: 0.00

Phlux Weight: 0.00

Amateur Weight: 0.70

Phlux Weight: 0.30

Amateur Weight: 0.30

Phlux Weight: 0.30

Amateur Weight: 0.70

Phlux Weight: 0.30

Amateur Weight: 0.85

Phlux Weight: 0.50

Amateur Weight: 0.50

Phlux Weight: 0.70

Amateur Weight: 0.00

Phlux Weight: 0.70

Amateur Weight: 0.30

Phlux Weight: 0.75

Amateur Weight: 0.75

Phlux Weight: 0.85

Amateur Weight: 0.30

New Prompt: An amateur photo of a 19-year-old Greek woman sitting on a couch in a living room that is dimly lit by a lamp with a warm light. She is wearing a green sweater and there is a bowl of popcorn next to her. She is smiling.

Phlux Weight: 0.00

Amateur Weight: 0.00

Phlux Weight: 0.00

Amateur Weight: 0.70

Phlux Weight: 0.30

Amateur Weight: 0.30

Phlux Weight: 0.30

Amateur Weight: 0.70

Phlux Weight: 0.30

Amateur Weight: 0.85

Phlux Weight: 0.50

Amateur Weight: 0.50

Phlux Weight: 0.70

Amateur Weight: 0.00

Phlux Weight: 0.70

Amateur Weight: 0.30

Phlux Weight: 0.75

Amateur Weight: 0.75

Phlux Weight: 0.85

Amateur Weight: 0.30

Conclusion

Both of these LoRAs improve on realism on their own. Amazingly, they each seem to fix each other's flaws when used correctly. Notice the raw Flux output is improved by adding either LoRA at 0.7 weight. Then that image is improved again by adding in the other LoRA at 0.3 weight. Using too much of either LoRA starts bringing out flaws with anatomy, textures, and massive popcorn (0.8 of each or more than 0.8 of either)

TLDR: Use Phlux at 0.70 and Amateur photo at 0.30 or vice-versa for best results. Tweak from there as different prompts/seeds yield different results. Sometimes 0.50 / 0.50 works too.

37

Comments