
This model is not intended to be used on it's own, although you could, I wouldn't expect great results doing that. Instead it's intended to be used as a pre-renderer to create the base image which you then upscale using your preferred SDXL model.
I find that while SDXL models produce the best quality, they tend to be very limited in their imagination, and produce a narrow range of images that all tend to look the same for a given prompt. SD1.5 was much better at producing a wider variety of image styles, but lacks the quality of SDXL.
By generating the initial image in SD1.5 and finishing it in SDXL you get the best of both worlds.
The example images shown here are meant to show off the results of this process, and are therefore provided without any inpainting or correction, and without any LORAs to give the most accurate and honest representation of the results.
Use the trigger word 'nvinkpunk' to invoke the unique artistic style shown here.
My recommended settings:
Step 1: Model: Halcyon Inkpunk, Steps: 24-34, Sampler: DDIM, CFG scale: 6-10 Size: 640x800,
Step 2: Select Hires upscale model: Halcyon v1.4a NSFW, Hires sampler: DPM++ 3M SDE Karras, Hires upscale: 2.7, Hires upscaler: 4x_NMKD-Siax_200k, Denoising strength: 0.4-0.5

描述:
训练词语: nvinkpunk
名称: halcyonNvinkpunk_v10.safetensors
大小 (KB): 2246054
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success
