Lea Seydoux (2014-2018) [FLUX.1D]版本v1.0(obsolete) (ID: 944480)

Lea Seydoux - photorealistic lora, works with FLUX1.dev, FP8, GGuf versions. Can be used with inpaint.

The LORA key-word in not mandatory, so you use either her name - "Lea Seydoux" or words as: "woman", "girl", "she"...

Due to the fact that Flux.1 dev model highly bloated with realitic images it might be hard to get non-realistic images using realistic LORAs, in that case you'll need to adjust CFG and Guidance.

Weight: 0.7-1.0 depending on your prompt and style.
GUIDANCE: 1.8-3.5 depending on your prompt and style [lower Guidance = more "creativity"].
CFG: 1.0-2.0 depending on your prompt and style, CFG at 1 means that model fully ignores negative prompt. [higher CFG = more prompt adherence, but at cost of x2 time generation].
Samplers: DEIS/UNI_PC_BH2/DPMPP_2M/IPNDM/DDIM - for more realistic skin texture or EULER - for smoother skin texture.
Schedulers: SGM_UNIFROM/BETA/DDIM_UNIFROM - for more realistic look and SIMPLE/NORMAL - for smoother and cartoonish look.

Less steps for close-up generations -> More steps for full-body shots. Usually [20->32].

My workflows, including one for loras - My Profile (openart.ai)

描述:

训练词语: LSeyd