city96
.
Can you try increasing the cfg value to 4 or 5? I've also noticed some differences between the implementation from diffusers with this repository. What I observed is that images tend to get quickly saturated with the SANA custom node compared to the diffusers pipeline.
Try adding more negative stuff like illustration, art, cg, toon, etc... With all this stuff excluded it looks pretty much photographic with some minor Limbs Issues (tm).
3d, 3d render, sculpture, figurine, effigy, cartoon, toon, cel shading, illustration, stylized art, digital art, digital painting, depth of field, blurry, low quality, ink art, artwork, painting, closeup, anime, drawing, digital painting, vector art, comic art, advertisement, tattoo, rags, cheap camera, low-angle pov, view from below, marvel, text,
and plus extra for a very specific use case testing out how much hulkish photo i can squeeze out of it due to anatomy bias on "muscle"
mutation, creepy, horror, anatomic atlas, overly detailed, body art, splotches, gore, anatomy illustration, bloody, scars, surgery, organs, dissection, anatomy study, zbrush, skeleton, concept art, monster, alien
Try adding more negative stuff like illustration, art, cg, toon, etc... With all this stuff excluded it looks pretty much photographic with some minor Limbs Issues (tm).
3d, 3d render, sculpture, figurine, effigy, cartoon, toon, cel shading, illustration, stylized art, digital art, digital painting, depth of field, blurry, low quality, ink art, artwork, painting, closeup, anime, drawing, digital painting, vector art, comic art, advertisement, tattoo, rags, cheap camera, low-angle pov, view from below, marvel, text,
and plus extra for a very specific use case testing out how much hulkish photo i can squeeze out of it due to anatomy bias on "muscle"
mutation, creepy, horror, anatomic atlas, overly detailed, body art, splotches, gore, anatomy illustration, bloody, scars, surgery, organs, dissection, anatomy study, zbrush, skeleton, concept art, monster, alien
Thanks, results seem better with your neg prompts.
How are people even getting usable results with this? Can someone spot and error in my workflow?
I've tried many prompts, nothing looks even close to real. Even a tree doesn't look real.
With this node: ComfyUI-Sana, and set pag to the default 2.0, the human anatomy seems better than your results(do not use the sana pag adjust node), although still far from perfect.