Search results for query: *

Hmmm interesting. Bet a little optimization could have reduced several hundred megs, or more. Bit of a pain every post is over a gig. edit: A lot of PDF's that could have optimization or be unoptimized and let lzma do a better job; jpegs that could be optimized and then leaving uncompressed in...

Another option is to concatenate them to a single file. A LOT easier if you use MinGW/CygWin or Linux. #Linux/Unix cat Archive* > Archive.7z #windows, i think this was how you do it copy /B Archive.zip.001+Archive.zip.002+Archive.zip.003 Archive.zip But i would certainly rely on 7zip first.

I wrote a script to do the same thing for epubs, stripping formatting and unneeded blocks, and do each paragraph separately. Compresses to about 200Mb per book. Unfortunately it's not consistent person output, which you can get around or not. If i find a way to make it consistent i'd be more...

I see several models that are promising, though using specific wording and Lora would help keep consistency; but only if it's already trained. Inpainting and other may work and just re-rolling until outfits or workable scenes may work too. I need to test, but i think you might be better using...

Could. Though i'd downloaded from Anne's Archive on stuff and found OceanOfPDF instances. So not sure. Honestly it's not the worst possible injection, but it is annoying when it breaks sentences or paragraphs and is on every page. This i also wrote like 2 years ago, but converted to make it...

Bonsai put out a 4B/8B 1bit model some months back. It was.... okay.... as an experiment and proof of concept i wasn't that thrilled. However they've recently put out a 27B 1bit model, which is just under 4Gb, along with a webui you can have a LLM on a DVD (Though the 20 minutes to copy to RAM...

If you were using 7zip, just 7z -x "MrDouble 2.zip.001" should have worked. you could also unzip it with the unzip command once concatenated. the test in it showed no errors when i ran it, thus it should extract just fine without 7zip at that point. Still long as it works it doesn't matter...

...same directory for it to extract. For unzip, i'll try a couple tests, but no promises. edit: Well that was fun. First test worked. cat MrDouble* > test.zip unzip -t test.zip No errors. So unzip then next should work just fine. A note, on the zip format especially for older splits, is...

Mhmm. If the options for how it generated the decisions for how it renders the output, one could make the voice consistent, or make several profiles for different characters. At this moment I'm mostly relying on KoboldCPP, which works with several models, though QwenTTS runs on the GPU so i'll...

GF has been listening to a lot of free audiobook stuff which are naturally AI TTS. So i wrote a script and i can manage to do something similar, though it is kinda mishmash round-robin as each paragraph is generated separately and then mashed together. The script itself isn't that complicated...

If anyone has any issues let me know.

This took way too long to get around to. Sorry... threads/trimming-epubs-oceanofpdf

...xdeltas should be usable to restore the original file (should you decompress the zip files) but should only worry about it if there's some major f*k-ups. edit: remembered and checked for invisible hyphens. That's added back in now. edit: go figure, forgot to see if it even ran. Fixed...

ComfyUI doesn't seem to like me much, whenever i mess with it it barfs. So GGUF models seem out for now. But yes i can run SDXL models, tends to take 5 minutes an image.

Mhmm. Don't quite have the hardware for some of the other models. Though found icerealistic which looks great for making KLRXO or other story/book covers, especially with hi-res fix...

Trying some hi-res fix tools built into A1111, and... i'm surprised at how much better the picture can be. Either from upscaling and then downscaling, or from just using a SD1.5 model and then getting 1024x1024 images. Fascinating...

Good news, i think a new model of 'realism engine' has some promising results.

I've been experimenting with putting a book/game into the metadata of an image. It doesn't like tags it doesn't recognize, but it works with ones it's familiar with. Aside from very small books, or say shoving NES ROMS into a cover image so you get the game with the image, it doesn't seem...

Believe it or not, the prompts are within the image as a comment (including the hash of the models, the seed, loras, etc); However it's not like it's a secret. Had some of the whipped images for #3, and hadn't gotten #4 done yet.

Well here's a few of the pornmaster's results, they do look photo realistic.