Hello everyone!
This is the first model I make based on Checkpoints Merge.
For this merge I did a lot of tests with different values, which I don't remember exactly. What I do know is the "recipe" for merge:
talmendoxlSDXL_v11Beta (by talmendo) + xl6HEPHAISTOSSD10XLSWF_v10 (by Afroman4peace) = MiaModel_v1
altxl_v30 (by altron) + MiaModel_v1 = MiaModel_v2
The idea of the merge is to get the maximum power from the models used until reaching a model that offers incredible results in almost all the generations carried out.
UPDATE!!! Model v3.0 recipe:
NightVision XL (by socalguitarist)
TalmendoXL (by talmendo)
GalaxyTimeMachine's (by galaxytimemachine)
Cherry Picker XL (by tkvier)
XL6 - HEPHAISTOS (by Afroman4peace)
AltXL (by altron)
sdxlbase_nsfw_faces (by sevenof9247)
WyvernMix - (by wier)
Positive prompt tips to get higher quality images:
Example 1:
(best quality), (masterpiece), (best lighting), (high detailed skin:1.0), detailed eyes, 8k uhd, dslr, soft lighting, best quality, film grain, Fujifilm XT3, ........ ............
Example 2:
detailed and realistic close portrait of ................... portrait photography, magical photography, dramatic lighting, photo realism, ultra-detailed, intimate portrait composition, Leica 50mm, f1. 4
Example 3:
masterpiece, best quality, (colorful), cinematic lighting, extremely detailed CG unity 8k wallpaper, ....................
As always, all your criticisms and advice will be taken into account, since they help me to continue learning and growing!
Do not forget to visit my guide to create Textual Inversion for SD 1.5
https://civarchive.com/articles/692/my-method-to-create-a-textual-inversion
If you like what I do and you would like to support me, you can do it with a coffee, link below ♥
Description
FAQ
Comments (8)
There isn't much models out there to merge with each other yet. I guess we're all dependent to that someone who has access to A100s or H100s to start training the base model with thousands of images. It's like starting from scratch with 1.5 but in the end it will be worth it.
Where can we find that thousand images dataset
That's right, there aren't many models yet, and creating a model from scratch requires too many resources, which in my case at least I don't have, that's why I opted for SDXL at least when performing merge tests.
It took about 3-4 months to get some decent 1.5 models. I think it'll be a few months before XL gets some decent models, but adoption rate is slow, people are too comfortable with 1.5. I myself am still using 1.5 models, dont find XL to be up to the quality of 1.5 models yet, but I definitely will switch instantly when a good XL model comes out.
@fox23vang226 I think that's it right there. A lot of people are too comfy and cozy with 1.5 and the slew of models and tools for it. It might take a little longer but I believe SDXL will eventually take over. Although, with how quickly SD went from 1.4 to 1.5, I'm wondering if were in a situation where StabilityAI might feel the need to tweak 1.0 just a little bit more to maybe a 1.1 version. Let's see how thing look in 6 months.
@fox23vang226 I think that's it right there. A lot of people are too comfy and cozy with 1.5 and the slew of models and tools for it. It might take a little longer but I believe SDXL will eventually take over. Although, with how quickly SD went from 1.4 to 1.5, I'm wondering if were in a situation where StabilityAI might feel the need to tweak 1.0 just a little bit more to maybe a 1.1 version. Let's see how thing look in 6 months.
Very cool model, I had a lot of fun playing with it. Not too flexible, doesn't always follow the prompt, but that's pretty much expected at this stage. I was skeptical about SDXL at the beginning, but now it's clear to me, that it has an enormous potential.
thank you very much for the review
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.