bemis
Low rep power
- Joined
- Mar 7, 2024
- Posts
- 79
- Rep Power
- 158
that picture of Nick in the visual novel screen was made using your LoRA in ComfyUI with one of the XL anime checkpoints. When I found it I was excited because it saves me some effort to prove some ideas, and I get white-pilled knowing somebody else had the same idea. I think once I can generate images that consistently look like Nick along with some other characters for foil, the short stories will almost write themselves. I might just start a thread asking for ideas when it's ready.Hey nice you found the LoRA I made! That’s sick that you’re working on an upgrade. The images from your visual novel could be great training data. I was able to get decent photorealistic images using the LoRA with the Realistic Vision XL model, but I could never get good animated images with other models. My theory is that feeding it some animated training data could help with that.
Your speech to text project looks dope too. I bet that’d be really useful for clippers looking for timestamps. I wonder how hard it’d be to feed that data to a chatbot and use it as a kind of wiki for AF viewpoints on people/events.
I know there is cozycaptions.com, I don't know the extent of it's capabilities but it seems like it's had a lot of thought and effort put into it. I'm kind of still feeling out what to do with my own system. My thought was give somebody the ability to search for a text string(s), return all instances of that, and let the user select which ones they want the video for, and then use ffmpeg or something to render those clips, and send the user a notification when they're staged for download somewhere. You could search for "good evening", concatenate all the files you get for it, and you've recreated that scene from his intro