This is my LoRA extraction of the Cosmos-Predict 2.5-2B. Using the Pre-trained and the Distilled models. Converted from pt to safetensor then cleaned during extraction.
Ready to merge.
CFG: 1.0
Steps: 8 - 15 Around 12 is good
Strength: ~1.0
Sampler: er_sde / euler_ancsetral
Scheduler: sgm_uniform / kl_optimal
v1-Refresh
Budgeted 64 dim. Better balance, lower file size.
v1
I have included 3 variations. 32 / 48 / 64. Up to you which you prefer. I would recommend the 32. Its what i used for all the image previews.
I do this for fun and to learn. Likes, comments and reactions are nice to see if my efforts are worth while and posting. I dont really care about the buzz or whatever but drop me a comment and like if you enjoy what Ive done. <3
Description
Initial Release
FAQ
Comments (6)
a very nice "default style"
Its great for merges :)
Is this great LORA the same as
https://civitai.com/models/2466415/cosmos-predict25-2b-base-distilled-extracted-dmd2-lora
Similar, its made from the same 2 models. I used a different method and still experimenting with various ways to improve.
@SpicyKatsu I'm yet to try your one but would you say your one is better as I'm currently using it with the official anima turbo Lora 2. Also, I notice you recommend rank 32. Is it better than rank 64 with this lora. Thank you.
In my testing, 32 had the best balance. I mainly posted all 3 if people wanted to play around or experiment with merging them. For generations you probably wont notice too much difference between the 3 but there will be some difference. It works best when paired/merged with something but is still good standalone.
The loras i made with merging this imo are better than this alone.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.