FLUX.1 Kontextで雑コラをリファインする
nomadoor.icon
FLUX.1 Kontextは目を持っているわけではないので、Subject-Driven Image Generationは出来ない
入力した画像の位置情報に引っ張られる点しかり、ただlatentを結合しただけ
ただし、超高性能なimage2imageモデルではあるので、先に生成させたいものたちをコラージュさせ、Subject-Driven Image Generationではなく、画像編集としてのタスクに置き換える
この機能をパワーアップしてくれるLoRAが出た
https://civitai.com/models/1780962/place-it-flux-kontext-loraPlace it Flux Kontext LoRA
https://civitai.com/models/1780880/fuse-it-kontext-dev-loraFuse it Kontext Dev LoRA
Transform the flat duck illustration into a realistic duck stuffed animal wearing a hat that the woman is naturally hugging while sitting on the bed
https://gyazo.com/5771f27e35298b919e3f216847214e6f
Add the yellow handbag to the woman so she is naturally carrying it, European street with old buildings
https://gyazo.com/be997efbbc0c0802513bfab1e8ebe585
polar bear wearing sunglasses sitting in a train seat drinking Coca-Cola, with jungle visible outside the window with motion blur
https://gyazo.com/9eec8fd66aa3ff0a286f05f9c1e78fb4
Convert the cartoon character in the jar into a realistic miniature figure submerged in water,wearing a snorkel mask and tube.
https://gyazo.com/967bb07db193bdc681c1f5528f99d537
https://gyazo.com/226a304940017fff3f34ab7c8ba93e6a
Refined collage with Flux Kontext.json
普通のFlux.1 Kontext (gguf)
https://gyazo.com/4555976febec3a524285fc7bf2cc1312
Refined collage with Flux Kontext_inpainting.json