For Flux

All images I experimented on ComfyUI. You can just simply use my images for nodes and workflows.

Parameters & guides:

Here are models that I used: flux-1dev, fp8_e4m3fn, t5xxl_fp16, and the original vae Flux

• Use Flux sentence-style prompts

• no trigger words required

• Lora weight: 0.95

• scheduler: Euler with dpmpp_2m

• for DualClipLoader: only place prompts in t5xxl box and leave clip_1 empty

• max_shift: 1.15

• base_shift: 0.50

• Steps: 25 or 50

• denoise: 1.00

• guidance: 3.5

Share your creations here and give some suggestions for whom you'd like me to make a LoRA model of.

for Pony

All images I experimented on A1111. You can just put my images on "PNG Info" on A1111 or any preferred tools.

Suggested Parameters & guides:

• Use any pony model, but I suggest using DucHaiten-Pony-RealV1.1 and CyberrealisticPony_v63.

• Trigger words:

  • No trigger words required. However, try adding "Chinese" in person prompt such as "teenager" >> "Korean teenager" can sometimes boost lora.

• Lora weight:

  • best at 0.10-1.00

• Sampler:

  • DPM++ SDE Karras or Euler a

• Resolution: 832x1216

• Steps: 18 - 30

Enjoy!