Please post your gens. Post them if they suck, post them if they’re great. I need to know how you’re using it so I can refine it properly.

1-22-35

Update i was able to *upload the model after probably 30 tries.

1-21-25

v0.98 - back to useability. give it a try. try it with clip skip 1 or 2. its a mix. give it a shot with 10 steps give it a shot with 80 steps. post your results.

*I've been trying to upload this model for a week. Different computers. Different browsers. No extensions. VPN. Always CORS policy exception. So I'll be uploading it via my iPhone which makes me zip it to upload it. I apologize for that. If there was another option like via google drive i could do that. i don't have issues uploading files to any other hosting service. I was finally able to upload the model, there's no fix so if anyone sees this and knows of something that I'm missing please help. its the number 1 reason i don't update models on here as often.

12-23-24

1.3 sd model refresh back to more of a semi-realistic look as the other model was burnt to bits. hopefully the next version will be new and improved and surpass 1.1

as far as .94 pony. i concur its trash as well. .9 is still better. back to the drawing board on that one.

10-31-24

Phantasmagoria v94 Pony

Happy Halloween.

This is a strange pony mix, half is trained with danbooru style tags, and half is trained with long-winded ai prompts that i crafted with Claude 3.5. The prompting of these tags includes director's names that often involved in horror-thriller films, the description of the subjects, with the lighting and shot type. Try it out and see if it works for you. i worked hard on the dataset but i don't really think it comes through. may need to train on vanilla sdxl. goal is to get a model that includes artist prompting, specific detail, and that will run on mobile hardware.

10-24-24

Sd v 1.2

Testing. Not official

062424 - some hours later...

Bad model removed. Upped good model. Don't forget to like and review and post your images.

062424

Currently broken- will get errors in automatic1111. Will upload and fill in the prompts and add more info in the next 24 hours. If you’re using draw things this will work but it seems off. I have the fixed version but I need to upload it via a desktop

—05/31/24–

Phantasmagoria v1.0 release.

The recommended values are ones shown in the example images. Avoid getting too verbose, use commas, and use the negative listed that starts with “BAD_IMAGE…”. *

Give it a shot and see if you like it. If you have any issues, content concerns, or want to just send me love or money please utilize the chat function or email me at [email protected]

—

• *all generations that you see are either done in automatic1111 with upscaling, or in drawthings ios app with native upscaler. no comfyui or post processing used albiet the title which I added with Picsart iOS app. ive included all the generation parameters in every image. most of them have issues but i want you to see what the model does. it can be helpful especially if you know photoshop to help clean up the images afterwards. getting this model close to being ready for a v1 release.