I've always liked Flux even though it is heavily and unnecessarily censored. Chroma is basically more what Flux should have been from the beginning. It is based on Flux Schnell and has all of the benefits of Flux without most of the Flux censorship.
GonzaLomo Chroma is a blend of Chroma 1 HD with a number of LoRAs, including some of my own. The primary goal was to try to retain Chroma's creativity while making it more consistent and stable.
The images I post with this model are created using my workflow which generates the image using GonzaLomo Chroma and then does a light refinement using GonzaLomo v6.0 Photo XL. The amount of refinement is configurable by changing the denoise value on the refiner's sampler.
Description
A little bit sharper image with better detail.
Recommended settings:
Steps: 8-14
Sampler: euler, dpmpp_2m
Scheduler: simple, beta
CFG: 1.0-1.3
FAQ
Comments (30)
I really REALLY REALLY like this @GBRX thank you so much for this, chroma prompt understanding of natural language is incredible =^.^= , I think you've improved it with your polish and that just makes it that much better . To anyone reading this whose hesitant to try Chroma this is the one to choose to try it :)
You are swiftly becoming a legend. Excellent workflows, models, images. Man- to be a tenth of how good you are. Thanks.
Thanks so much!
Thank you for your nice work for all of us
OMGbrx! I don't know what you did in V3.0, but my lora and workflow LOVE IT!!! Thank you so much for making and sharing... Long live Chroma... 💗
Glad someone is furthering along chroma developement,Awesome job thanks!!
In one word ridiculous model....ridiculous good i mean :).
It does everything,great prompt adherence, great quality. You think it and you can make it.
I hope you find a way to even make this beter, perhaps quality wise.
Good job.
Am i blind or a are the faces basically looking all the same? probably an issue regarding turbo/flash mode?
I don't see an issue with facial diversity. As with most models, it has a default type of face if you don't supply details in the prompt.
I have the same issue, feels like illustrious in that regard, face diversity is very low.
@mark_sausage439 Hmm, I just looked through all the showcase images and don't see a same-face issue. That said, the number of images I used to train on Chroma 1 HD is about 100. I think the dataset was pretty diverse but I'll have another look.
So confused. All the workflows (images) Don't even use the checkpoint I downloaded. SMH. The one called gonzalomoChroma_v30.
I'm not sure what you're referring to. What "workflows (images)"?
@GBRX The ones attached to this page at the top. When i drag any of those images into comfyui
@saeroner561184 I can assure you they use the same binary checkpoint. The name might be slightly different because Civit sometimes changes it from the filename that I have locally.
Can you make smaller v3 in Q2 or Q4 GGUF version?
I'll see if I can do a Q4_K_M soon.
@GBRX Still waiting for q4_K_M for v3.
@GBRX would love a gguf version!
@physis I've got a Q4_K_M ready to go but the site is pretty borked atm and none of the images are getting rated.
It'll be published in an hour.
@GBRX Thank you so much!!
Perfect model, it allows to make a lot of interesting poses. The best solution for two or more figures interaction. Thank you very much!
This is the best low-step model for Chroma. I hope it will continue to develop further because the potential of Chroma has not yet been revealed, also due to the terrible official Chroma1-Flash model, which is easier to throw out than to make it produce good results.
Thanks!
That's a really really great model, thanks for that. May I ask how you got it to 8-14 steps? With the Heun-Lora? And if so, which version? 256? 128?
Thanks! I used chroma-flash-heun_r128-fp32. I did try the 256 version but didn't see much difference, especially with the size increase.
Finally decent chroma model! Thx!
I'd like to express my deepest gratitude to the creator of this model. It's the best thing out there right now, seriously. The prompts are easy to control, and you can create any kind of scenario. The image quality is amazing. It's a shame the parent model has the same issue with the hands, but that's not a problem considering the generation speed.