JARDABEREZA: jo, ale ten ma ten inpainting pry delany docela neintuitivne. Ale asi se do toho budu muset ponorit teda

predem se omlouvam na zacatecnicky dotaz, ale jaky software byste mi doporucili na pouziti lokalnich modelu pro fotorealisticky inpainting? Idealne aby to bezelo na Linuxu. Zkousel jsem Fooocus, ale tam ni nektere modely nejdou nacist,a pak Invoke AI, ale to dela hrozny kraviny (a/nebo s nim mozna neumim). Hardwarove nejsem omezeny, jde mi o co nejlepsi vysledek.

JARDABEREZA: specialne na image edit je imho nejlepsi qwen-image-2.1 (na cokoli ostatni uz ne tak moc)
u krea2 a moss to jde take kdyz pouzijes identity-edit-lora.

ICE: Ale z toho, co vím, tak nemají nativní image edit.

If you're starting from nothing, the VRAM is the thing that decides everything. 8GB runs 512x512 fine but gets tight once you try DreamBooth or upscaling, and switching models all the time eats the RAM more than people expect. I'd look at the card first and forget the rest of the build until then. What size images do you actually want to make?

nekdo postrehy k Stable Diffusion 3.5? chci si SD znovu po dlouhe dobe zkouset, co je ted best workflow?

aha dekuji vsem

//např. online tv a youtube