this is impressive, but there are a few issues from a practical perspective
#1
by
fanmcleod
- opened
- there are checkpoints people prefer to use because the quality of the output is significantly higher than yours
- this can't let others use those better checkpoints
- this isn't a controlnet
- this doesn't work alongside controlnet from what I can tell
- this implementation doesn't let loras help to increase the quality
- this only works for frontal shots and doesn't for side perspectives
Something for issue 2, If running locally.
Modify /fastcomposer/fastcomposer/utils.py > Find this line: default="runwayml/stable-diffusion-v1-5",
Change to diffusers model location.