A close-up photo taken from a presentation in a conference room shows a slide titled "AnimateDiff" which explains how to transform domain-specific text-to-image (T2I) models into text-to-video (T2V) models. The slide mentions that domain-specific (personalized) models are widely available for image generation, with finetuning methodologies like LoRA and DreamBooth. It also references communities such as Hugging Face and CivitAI. The core task outlined is to convert these image models into video models without specific finetuning. The slide includes an example interface displaying various character images, presumably generated by these models. The photo captures the bottom half of the projector screen, part of the conference room wall, and the top of an attendee's head. The authors of the work cited on the slide are Guo et al., with reference to an article from arXiv in 2023. The work appears to be led or associated with Mike Shou, NUS.
Text transcribed from the image:
Animate Diff
Transform domain-specific T21 models to T2V models
•
•
Domain-specific (personalized) models are widely available for image
•
Domain-specific finetuning methodologies: LORA, DreamBooth...
Communities: Hugging Face, CivitAl...
Task: turn these image models into T2V models, without specific finetuning
CIVITAI
MOST DOWNLOADED
ALL DIMACTER LIKE CELEBRITY CONCEPT GOTHING
TOOL BUILDINGS VEHICLE OECTS ANIMAL ACTION ASSETS
THX
RealCARTOON
3D,
ON OF LIKE PAR
Cyfarfalite
Cute TERR style)
OLE OF A F
OLE OF AIR 2
Guo et al., "Animate Diff: Animate Your Personalized Text-to-Image Diffusion Models without Specific Tuning," arXiv 2023.
QA
MORING FASC
MONTH 9
[Ouickpoint) Tolly
Q10 On 1 ta
RealCARTOON
Copyright Mike Shou, NUS
1: