HassanBlend 1.5 progress
Added 2022-12-14 11:12:59 +0000 UTCHey all,
As you know from the previous update, the finetuning that I performed over 4-5 days last time wasn't efficient enough, mostly down to captioning. Since that last update what I've been doing is:
- Manually writing captions for batches of the dataset, there's almost 7k images in total so I'm doing batches of 3-500 at a time
- Finetuning each batch one at a time, so it's an iterative approach to finetuning the model instead of doing it all in one go
- Testing each iterative version and rinse and repeat until I'm satisfied
Here's a teaser on the current progress, these are txt2img only, no img2img.
Note: As mentioned last time, patrons will gain early access based on tier and then the model will be released free for all as normal.
Ultimate gain first access, Supershot a day or two later, followed by Boosted and then released free
More images will be shared as samples in the discord too
https://discord.gg/sdmodelers
