I’ve been using qwen 3.5 abliterated. I was wondering if it’s still the best cheap to run but still decently high quality model out right now (kinda like what Z image or Klein are in the image gen space). I don’t use it to code, just chat and creative writing and sometimes flux prompting
I’m trying to import a few Loras for Klein. Most work fine, but sometimes I’ll try to import one and I’ll get an “incompatible” error despite me manually choosing the correct model in the import menu. Has anyone else had this problem?
From what I have tried WAN is still off limits, but I haven’t tried ltx yet. Has anyone generated with it and got decent times using similar specs?
I've seen a few guides on how to do masked inpainting with z image turbo for comfy, but they never seem to run well on my mac. ZIT runs great on my mac for t2i in draw things, but I don't know how to do inpainting with it. The ui is a little harder to understand than comfy for that purpose. Has anyone done masked inpainting with z image in draw things before and is willing to share any tips?
I’ve noticed over the last year or so that the image2image scene has been dominated by full image edit models like Qwen, Kontext, Klein. I still prefer to do traditional mask based inpainting instead of feeding the whole image into the model and it changing every pixel. I’ve been using sd1.5 and sdxl models for this, but you can tell they are getting old. Skin looks kind of plasticy, hands look like sd hands obviously. Are there any modern models that do inpainting but have the insane photorealism performance that z image or flux models have? I’m open to custom workflows that use models that aren’t made specifically for inpainting if that’s the only option.