Lecture 5 (IDK)
Réviser ces flashcardsGratuit · répétition espacée · iPhone, Android, Mac et web
Toutes les cartes
1
What is the goal of AI video generation?
To create video sequences from various inputs, most commonly text prompts or images.
2
Which early (2022) model focused on generating longer videos from sequential prompts?
Phenaki
3
What is a major hurdle in AI video generation mentioned as a 'Core Challenge'?
Maintaining temporal consistency (ensuring characters, objects, and styles remain coherent).
4
What technique allows changing the text prompt during generation to evolve the scene smoothly?
Prompt Travel
5
What technique adds conditional control to diffusion models using inputs like human pose estimations, depth maps, or Canny edges?
ControlNet
6
What is IP-Adapter?
A lightweight neural network adapter that enables image prompting by guiding generation with a reference image.
7
How does IP-Adapter work?
By injecting image features into the diffusion model's cross-attention layers.
8
What is LoRA (Low-Rank Adaptation)?
An efficient fine-tuning method to adapt large pre-trained models to specific characters or styles without full retraining.
9
What is the primary purpose of 'Character Sheets' in the context of LoRA?
As training data for creating a character-specific LoRA.
10
What is the focus of specialized models like 'Animate Anyone'?
Animating a character from a reference image based on a driving pose video.
11
What component does the WAN model use to compress video/images into a latent space and decode them back?
Video Autoencoder (VAE)
12
What component does WAN use for the core denoising process in the latent space?
Diffusion Transformer (DiT) blocks
13
What architecture component does a Diffusion Transformer (DiT) primarily replace compared to earlier diffusion models?
The convolutional U-Net structure.
14
What does the 'Mixture of Experts' (MoE) approach in WAN 2.2 involve?
Using different expert models or components specialized for different stages (e.g., high vs. low noise levels).
15
Name one potential negative application (misuse) of generative AI video.
Disinformation (or deepfake commodification, scams, harassment).
16
How does Prompt Travel allow for intentional changes during video generation?
By modifying the text embedding used for conditioning the diffusion model at different specified frame numbers.
Plus de flashcards
Handball im Schulsport (Staatsexamen)AllemandPaulRohstoffe und FördermethodenAllemandluisS. 206 rechtsLangues · Italien → Allemandsalina_svlEspañol - Unidad 7Langues · Espagnol → HongroisRékaE10-1 VocabLangues · Allemand → Anglaisemma.beelChapter 16 - Corporate OperationÉconomie et gestion · AnglaiswowiepowweatherAnglaisanya€mito y leyendaEspagnolEvyRUSS 2340- Test 1Langues · Latin → RusseFearTheTrixPsychology Exam 3 (Lessons)Psychologie et société · AnglaisQuavoNursing Assessment: Neuro and Lymph NodesAnglaisMatthewCHAW 1 - Ancient History FiguresHistoire et lettres · AnglaisAylee
Révise ces decks dans l’app
Gratuit sur iPhone, Android, Mac et le web. Ajoute n’importe quel deck à ton compte en un tap.
