Testing the Best Flux Models and Workflows for Interior Design and Architecture

[@bgtubber]

Really cool demo! Thanks for taking from your time to show us all of the cool things we can do with Invoke.

[@DarthChrisB]

"Won't run on a Game Boy Advance" - *Laughs in Super Mario 64*

[@tomschuelke7955]

once again nice stuff... first question that apears for me.. the flux Ip Adapter... is it allready for 5.5 in the download section available, or do i have to download it somewhere.. I Think the example shows very well, where Flux already excels very fine. What i suspect wont work, is what "real architects"... not interior architects , are aming for.. big architecture, like 100 store office building headquaters, and again when you have to put in sketches and screenshots of big cad plans. I suspect the reason for this is just, that the training data focuses usualy on real world images taken from usual people or at last interiior photographs. But i think over time all areas of this will improve.. And yes.. We all love your product

[@rodrigop.9071]

It is possible to use Flux NF4 in Invoke?

[@agressor7905]

Yes, it works very well

[@CharlesMcGill-et8un]

@@agressor7905 How did you get it to work? I already added the model but it doesn't allow me to generate images. What am I missing? Thanks.

[@agressor7905]

@@CharlesMcGill-et8un how much VRAM and RAM you have? What is your videocard?

[@TheCritterPeddler]

I have a 12GB VRAM GPU but it is only using "6 point something" during generations. what do I need to add to the invoke yaml file in user settings to make it use more? the configuration tutorial in github mentions a vram setting but it's not clear as in how to write it for it to take effect. I remeber Invoke used to be able to run with all of my VRAM, in previous versions. is this no longer possible?

[@agressor7905]

my method for generating interiors. At first generate some pictures very quickly with any SDXL model (it hallucinates well and is fast), then Hires.fix x1.5-2 with sdxl, then generate the selected picture through img-to-img in Flux (it fixes everything, because it is very accurate). all this is done in two Forge tabs. I would do it in invoke, but there is no hires.fix.

[@andybroadbentphotography5024]

really struggling getting photo realism with trained character LoRas in Invoke, I have used same LoRas in Fooocus and getting more accurate faithful likenesses could you do a vid on this, I have played with LoRa strength, the scheduler type and settings and tend to get deformed faces (mainly lips) and if I try and inpatient on face to try and improve the resolution drops - looks like huge pixels in unpainted areas, I know it it will a settings thing but tearing my hair out

[@invokeai]

Hop into our Discord channel and the community can help troubleshoot there.

[@meepo1862]

can invoke run huanyuan video?

[@doyen86]

Yea, what would love to switch over to this but it takes 5 minutes to render an image on 12gb VR and only takes 40 secs on ComfyUI using the same fp8 model. What gives?

[@invokeai]

Check out our 5.6 release later this week. We'll have a low VRAM mode available for lower performance GPUs. It's in beta testing right now.

[@TheRoninteam]

I dont feel it is better for interior design, its kind of lame I guess