In-Context LoRA

https://gyazo.com/509b1e7f6ba8c02707a293e10dc44dd2

https://github.com/ali-vilab/In-Context-LoRAali-vilab/In-Context-LoRA

https://arxiv.org/abs/2410.23775In-Context LoRA for Diffusion Transformers

Group Diffusion Transformersや【Flux.1-Dev】プロンプトのみで一貫性のあるスプライトシートを生成するでわかるように、DiTは元々、複数枚に渡って文脈に一貫性のある画像を作ることができる。と仮定する

一貫性のあるマルチパネル画像と、複数のパネルをまとめて説明するキャプションを付けたデータセットでLoRAを作ることで、この特性を安定して引き出す

e.g.

portrait-photography.safetensors

https://gyazo.com/be792a320f080e6c880900d6e9c52068

font-design.safetensors

https://gyazo.com/935329eda3ee4dae0465cd464aaa8e63

https://huggingface.co/ali-vilab/In-Context-LoRAali-vilab/In-Context-LoRA

https://github.com/ali-vilab/In-Context-LoRA#model-zoo推奨解像度&プロンプト例

https://gyazo.com/eafe99f6567dae4db4f634faab6e7d7d

Film-Storyboard.json

https://gyazo.com/4e737fdc82b57dd96a85639fa16ab49c

code:prompt

MOVIE-SHOTS Set in the early 2010s, this inspiring tale of growth follows <Anna>, a 30-something woman whose life takes an unexpected turn in the world of fashion. SCENE-1 captures <Anna> from outside the restaurant window as she works, her appearance unkempt with disheveled hair, clearly unconcerned with her looks, SCENE-2 leading to a life-changing moment when an elderly man in a sleek suit, dining as a customer, approaches her with an invitation to join a fashion brand, SCENE-3 portraying <Anna> as the same man mentors her on dressing with sophistication, gradually transforming her style and confidence, SCENE-4 concluding with <Anna> in Paris, commanding the backstage of a fashion show, confidently giving direction to models and junior staff, fully embodying her new role.

サンプルをテンプレートとして、こんなシナリオに書き直してくれってChatGPTに投げると作ってくれるnomadoor.icon