Update 24.03:
The new version now achieves significantly better image quality, and thanks to Alimama Turbo, it all happens in just 4 steps. I've also switched the Prompt LLM from Florence2 to LLaVA via a local OLLama, as this setup works slightly better too.
Comfyui Workflow for the ComfyUI Gemini Flash nodes.
First, the image is generated with Gemini, then it goes through a Flux upscaler.
You can choose between Fast and Dev modes.
Dev looks better and works well with slightly higher denoise values, while "Fast" is… well, faster ;)
Installation:
Get your free API key from Google AI Studio:
Visit Google AI Studio
Log in with your Google account
Click on "Get API key" or go to settings
Create a new API key
Copy the API key for use in config.json
Set up your API key in the config.json file (will be created automatically on first run)
https://github.com/ShmuelRonen/ComfyUI-Gemini_Flash_2.0_Exp?tab=readme-ov-file
Description
FAQ
Comments (15)
Thank you for this workflow. I didn't know that you can generate pictures with the experimental version. It's a nice gimmick.
It's a bit like looking into the future. Of course, the multimodal models still have many bugs and the output quality is still very poor. But it's pretty crazy what the model can do.
For some reason, this hangs up the entire ComfyUI...can't zoom out, zoom in, switching to another workflow doesn't work. Anyone else facing this issue?
yep, straight to trash, breaks comfy, thanks for the effort OP , but it's unusable
@sacrificegoat154 Do you get any kind of error message?
@denrakeiw No, none at all. No error message in browser or terminal.The RAM/VRAM are fine(not overloaded). If I close this workflow, refresh the browser, then switching to another workflow works fine. Let me know if you want any more feedback :).
no error, so basically what happens is the workflow loads in, way below where it should be, when you start zooming out it will freeze when it hit one of the nodes just below the top big node, can not be sure which one. It's fine tho, idk why it won't work for some, maybe we just have something outdated. Or we could have a higher security level on comfy that doesn't let one of the sepcific nodes load. Hope all who can load it properly enjoy using it tho. I did find what I ultimately was looking for in another workflow. I just kinda liked this one better from the looks of it lol. But if it doesn't work I don't wanna waste too much time on it. No big deal. Also still not the worst, after this one I downloaded one that straight up broke comfyui after downloading the missing nodes ,so had to delete and use my backup LMAO, so your workflow is definitely fine XD
here is a bit of an update, https://civitai.com/models/1404390/wan21-fun-control-workflows-nativewrapper?modelVersionId=1587476 this workflow does the same thing. Freezes and can't do shit, no error, nothing, maybe there is some node in both that is the same ? there is a group in it called FLUX first frame controlnet, idk wtf is in there but even when I am trying to remove the nodes ,it keeps freezing up.
Can you please provide the link to the highresfix lora?
My Ollama Describer always outcomes with 'name' :(
Have you installed an Ollama server?
https://ollama.com/
@denrakeiw yes
@Santaonholidays also downloaded the llava model in ollama ?
@denrakeiw Yes