Basic workflow for Krea 2 photorealism. Results speak for themselves.
Text to Image w/Basic Latent Upscale.
er_sde/beta, 8-9 steps for text-to-image
exp_heun_2_x0/sgm_uniform, 3 steps for latent upscale
Things you need:
Model:
Text Encoder:
qwen3vl_4b_fp8_scaled.safetensors - Load with CLIPLoader, type
krea2:
VAE:
Nodes:
Like it? You can buy me a Ko-Fi (I'll pretend it's a beer) here:
https://ko-fi.com/raymondluxuryyachtrly
Thanks!
Description
FAQ
Comments (11)
Thanks for this. I've enjoyed your LoRAs. I recently switched to using Comfy Desktop and your workflows are very helpful.
Glad to hear!
appreciate this MAN!
Looks awesome but unfortunately I'm kind of a technical idiot. I can't figure out what the hell is going on with the sesqui node or how to download/install it. Any tips would be appreciated.
In comfyui, with the workflow open, click on the "manager" button at the top of the canvas, then "Install missing custom nodes". It should find it and you can install it from there.
If it's not found, instead of clicking on "install missing custom nodes", click "custom nodes manager" and search for "sesqui" and install it.
I confirm it's not available in ComfyUI. Nodes Manager --> All Extension --> Nodes = "sesqui" gives "No results found that match your search"
Auto install of missing nodes doesn't find it either.
@jelesi It's there for me, but I've enabled the legacy manager. The link to the node is also in the workflow itself. You can follow the instructions on that page to install it manually.
Short version: Open a command drop in your custom_nodes folder and type:
git clone https://github.com/LoganBooker/SesquiLSR.git
Then restart comfyui
@RaymondLuxuryYacht I using Comfyui Desktop 1.0.38. "Sesqui" is not available in the nodes manager. Manual install in the custom_nodes folder via "git clone https://github.com/LoganBooker/SesquiLSR.git" downloads the repo, but still shows "Missing Node Pack" in Comfyui (after restart). Any suggestion what could resolve this much appreciated.
Nice simple workflow. Would qwen3vl_4b_int8_convrot be a better match for the int8 model?
Not necessarily better, just more efficient. Should work fine!