Please ComfyUI pipeline
This could perfectly work in ComfyUI pipeline using DynamicVRAM-which is Unified Memory and VRAM not important anymore, the system RAM is. Ive tested RTX4070 16Gb on full Kandinsky 5 Pro diffuser model (76Gb) and it perfectly works in Comfy DynamicVRAM on consumer PC with enough system Ram. Moreover i compared and 4070 equal by generation speed to 3090(in Xformers). In such Unified memory the power and speed of GPU important now than memory.
About Flash attention, i would prefer more Xformers than that or even PyTorch, because such very easy to install (compiling Flashattention every time in new fresh system is really several hours wasted time) and ive tested on model above that Torch generation really slow, after installing on top of it Xformers it improved speed radically (~2 hours vs 40 minutes (full 50 steps in full model).
It's officially now supported in ComfyUI! In templates section.
I would only want if full model could be packed as diffuser file and used there. Like Kindinsky diffuser in 1 file which perfectly works in workflow. (i've tried your diffusers in form of shards and its not loading, Comfy dont understand that file).
Quality is very good, this is new type of "intelligent" editors - which understand every piece of picture, like in example below (turn every tree leaf color to autumn), others older type of image editors produce a mess on such prompts. Comparing it to other good intelligent editor from JoyAi Image Edit, Bernini preserve original details better, maybe less fantasy-so prompt need maybe more work here (JoyAi on same prompt produce more creative output, Bernini produce very straight total execution).
Bernini-R in latest ComfyUI (high fp16, low fp8 models)
JoyAi Image Edit (~30Gb model as i remember)
I havent installed any Flash attention, just using xformers attention.
Its basically like talking with Ai bot which drawing art, but which is quite "lazy".
There's only 2 ways with lazy LLMs - convince Ai to make better work or describe each part, portion, detail thoroughly. First way much shorter and faster. Prompt looking ridiculous in that way but it works, women could be very pro in such prompting.
