Please react to my videos to build rlhf dataset: https://civarchive.com/user/ntc/images?sort=Most+Reactions
Usage: add lora between 0 and 2.0
Use the trigger word smile
Trained in 2000 steps with:
"#:0.41|~smile:0.2|^smile:0.05"
https://github.com/ntc-ai/conceptmod
Training now optimized to run 4x faster. 1/2 of that was a bug with gradient accumulation not working,
Tutorial on how to train your own (updated):
https://civarchive.com/models/50070/mania-concept-trained-without-data-tutorial
Words are the means by which we shape the world, and the power of names is the power to shape the reality.
Description
FAQ
Comments (7)
would you kindly do an animation of marvin from hgttg with this lora? please make him smile :)
"I have the sort of feeling that any minute now I'm going to make a sudden, abrupt noise and then just fall apart like a broken toy. But still, if you want me to animate a character, I suppose I can try. Just don't blame me if it turns out to be depressingly tedious." - gpt4 as marvin
I just love these "with out data" loras collection 😍, I wander if you can manage to create movements concepts like "translation" what would rly be a step forward to in loRA based animations.
Great concept !
Can you show the full prompt of the first render, with the greatest smile ?
ok what am I doing wrong? Why cant I get the animation ? I have added it to the correct Lora folder, use the keyword - but its just a static image?
Hey you might be the first to actually get this far so thanks for the question. If the lora is not being applied correctly, or with an incorrect -e flag, the image could be static. There might be logs in a111.
Here is the exact command I'm using for the smile animations.
python3 lora_anim.py -s -0.0 -e 2.0 -l "smile" -lp ", smile" -np "nipples, weird image." -n 32 -sd 7 -m 6
Some other helpful debugging tricks:
- To debug whats being sent to a111, you can add a print statement to generate_images in lora_anim .py
- Run the prompts manually on a111 webui to make sure its all working.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.