Type | Workflows |
Stats | 422 0 |
Reviews | (10) |
Published | Jun 11, 2024 |
Base Model | |
Hash | AutoV2 E3FBEFFEA5 |
This uses a load image node hooked up to an LLM for an Image to Prompt node and it's asked what the picture is and to create a SDXL prompt from it. That's then fed to a second LLM which is asked to take the prompt and improve it. There are also manual positive and negative prompt boxes for trigger words and control on the output.
This all goes to a Group node which contains AYS, PAG and Auto CFG, then a face detailer, an upscale by model and finished with SUPIR. I've got the upscale on 2.5x as that's as much as my 2080Ti can manage, but you can raise this if you have a card with more VRAM.
It can be a bit random and chaotic (personally I quite like that), but can be good to get some inspiration and ideas for prompting.
For some reason it won't let me add other resources to the example images, so for those interested the main Checkpoints used for them were ZavyChromaXL and Proteus RunDiffusion.
LoRAs used were:
A rainbow exploding with joy - sliders
80s punk clothing - sliders
Cosmic Rainbow Style [SDXL]
Dissolve Style [LoRA 1.5+SDXL]