All versions require PickScore nodes to be installed.
Gemini 2.0 version also requires Gemini 2.0 Flash to be installed.
All versions that are using Gemini 2.0 are updated and are also using system prompt node that can be found in the creepy_nodes pack on github and in ComfyUI manage.
NOTE: If you have an issue loading the Gemini node, try to pip install sounddevice
This is a fully automatic workflow, mix the Detailer Deamon and Flux Redux, as well as Flux Dev and Stable Diffusion 3.5 Turbo.
I have included 2 nodes to clear VRAM/Cache at strategic places, but I can't say what the minimum system requirements might be. I have 12GB VRAM and 32GM system RAM, and the workflow is working well for me.
1 generation creates 5 images, of which one is upscaled, ran through detailer and finished with Stable Diffusion 3.5 Turbo.
If you have missing nodes, just go in tho manager and install them from there.
Upload an image and Florence will extract a prompt from it. Redux will make sure the images keep the style/theme and Detail Deamon will produce 4 images with variation on every run.
The images are then evaluated by Scorepicker and the best of the 4 images are choosen (the other 3 are also saved).
The choosen image will go through an upscaler. And after that the workflow changes the checkpoint from Flux dev to Stable diffusion 3.5 Turbo. With the SD 3.5 checkpoint the final image then go through a face detailer and finally through Ksampler with low denoise (0.15-0.3).
Description
This is the same workflow as Detailer Deamon - Redux, but with Gemini through API instead of Florence. There's also a Reactor node that can be turned on and off easily.
FAQ
Comments (2)
Instructions would be appreciated (detailerDeamonRedux_gemini20), but after installing this flow on my ComfyUI 0.3.26, missing nodes caused it to break. The ComfyUI-easy-use node no longer works since then!
Repairing ComfyUI is impossible. I had to restart a complete installation from the beginning with all node. Hours were lost trying to make this workflow work!
And after struggling to make it work, I failed with the following error:
"ZuellniPickScoreLoader
yuvalkirstain/PickScore_v1 does not appear to have a file named pytorch_model.bin, model.safetensors, tf_model.h5, model.ckpt, or flax_model.msgpack."
My opinion: Avoid this workflow, as it’s built with outdated models that break your up-to-date installation.
I'm literally running this workflow right now.
Instructions for Pickscore: https://github.com/Zuellni/ComfyUI-PickScore-Nodes
Instructions for Gemini: https://github.com/ShmuelRonen/ComfyUI-Gemini_Flash_2.0_Exp