different image in forge and swarm

[kalle07]

again ...
Maybe you discuss the topic on Discord ?!? (i dont)
Unfortunately I can't do it in compfy with a workflow!

so first
Forge

swarmui

(meta-data inside PNG)
i only can make an x/y/z plot in forge

so I changed the cfg and the cfg scale slightly

so for me its like you have in forge and swarm a different model ?!?

[kwal559]

not sure purpose of generating same pic twice, but, i get bad images at distilled CFG 0.1-1.5.. However 0.0 seems to generate great images. I think this sliding scale may be implemented differently per user interface.. InvokeAI i believe shows 2.0 - 7.0 (or there about)

[mcmonkey4eva]

Distilled CFG doesn't exist, it's a mislabeling of Flux Guidance Scale, and it's a direct parameter input to the model. It's possible invokeai does something weird, but most UIs just take the number you give and feeds it into the model. They recommend ~3.5 as the default, but lower values in the range of ~2 can be great. <= 1 is chaotic and shouldn't work, that's asking the model to given invalid output.

[kalle07]

it works well for an SDXL and a Pony model ... same parameters same image

hmmm, why not FLUX ?