image gen
I’ve had a lot of people ask me about how I gen my images, so here we are.
- Tools: ComfyUI
- Resources: Danbooru Tag Wiki
- My CivitAI Profile: I include my prompts + models + LoRAs there.
- Models: Antyhing by Sexiam or MadVlad
frequently used tags
Try mixing and matching these to see what works for you. I usually use either the general or painterly tag set, and then include the lighting and artist tags as desired.
### general
masterpiece, best quality, highres, good quality, very aesthetic, absurdres, newest, 8K, blurred background, realistic,
### painterly
impasto, oil painting, painterly, waterccolor, very aesthetic, sketch, lineart, blurred background, realistic, absurdres, masterpiece, best quality, amazing quality, very awa, very aesthetic,
### lighting
shadows, backlighting, dim,
### artist tags
(artist:96yottea), (artist:748cm), (artist:dokimdoki),
### negative
low quality, worst quality, worst detail, normal quality, jpeg artifacts, watermark, bad anatomy, bad hands, distorted, ugly,
sampler settings
steps: 25
cfg: 5.0
sampler: DPM++ 2M (dpmpp_2m)
scheduler: Karras
second ksampler: Same as above, but with 0.40 denoise and a different seed
upscaling and detailers:
I have nodes that sharpen the image and add a bit of contrast before doing a Remacri 4x upscale. It’s a bit jank, but I use lanczos to scale it down by half before detailing. After upscaling, I use Body, Face, Eyes, Hands detailers, but you can try Hires Fix if detailers are too heavy for your GPU.