Este modelo es el resultado de un merge experimental entre CyberRealisticPony_v90Alt1 y NevolutionSDXL_v2Base, dos modelos SDXL con enfoques muy distintos pero altamente complementarios.
VAE no incluido, use SDXL_vae
La idea detrás de esta fusión fue combinar la precisión en rostros y figuras humanas de CyberRealisticPony con la potencia creativa y riqueza fantástica de NevolutionSDXL. El objetivo es ofrecer un modelo versátil capaz de generar tanto escenas realistas detalladas como composiciones oníricas, surrealistas o épicas sin sacrificar calidad en ninguno de los dos extremos.
Es mi primer merge, así que toda crítica constructiva y feedback son bienvenidos. Espero que lo prueben, experimenten con él y, sobre todo, que les inspire tanto como a mí al crearlo.
¡Gracias por pasarse y feliz generación!
This model is the result of an experimental merge between CyberRealisticPony_v90Alt1 and NevolutionSDXL_v2Base, two SDXL models with very different but complementary strengths.
The VAE is not baked, use SDXL_vae
The goal was to combine the highly accurate human figures and facial details of CyberRealisticPony with the fantastical creativity and visual richness of NevolutionSDXL. The result is a versatile tool capable of producing both lifelike portraits and dreamlike, surreal, or epic scenes—without compromising on detail.
This is my first custom merge, so I'm open to feedback, impressions, and suggestions. I hope you enjoy using it as much as I enjoyed creating it.
Thanks for checking it out, and happy generating!
Description
FAQ
Comments (22)
Hi everyone. I'm trying to refine the model to incorporate more fantasy elements without losing the photorealistic quality, but every time I try it, it either loses quality or ends up with broken images. I welcome suggestions; I'm new to model mixing.
Maybe try mixing in a reality checkpoint from @Sateluco or @homoludens (or some of each). They both have some really good realistic-fantasy models in both Pony and Illustrious. Or save yourself a step and just mix in some of HL's XenEro v1.8-SatMix, which is an amazing model of that genre.
Also, maybe bake-in an SDXL VAE or at least make a note that one is needed with your model. My results were all washed-out until I manually loaded one. Just some ideas/suggestions, best of luck!
@AFD_0 Elemental merge (MBW mode)
Build a dataset. Redraw it with higher details with each new checkpoint you produced. Each improved checkpoint you have made can be applied to the same dataset to give it more details. I'm using ComfyUI to improve details of my base dataset. With each iteration level of details gets higher and higher.
Topics you need to learn:
1. OneTrainer + Masked Training
2. ComfyUI + Ultimate SD Upscale
3. ComfyUI + WD1.4 Tagger + Qwen2 VL Instruct
4. ComfyUI + MBW Merge
@homoludens Did Sateluco rename or delete some things or is that the same as SatPony-Reality v1/IOR? Was thinking there was a checkpoint named SatPony-IOR with that versioning, but maybe I'm misremembering.
@AFD_0 IOR outdated. There was IOR2, but then Illostrious experimental V3 released. I'm focused on adding details into my datasets and don't check if any improvements added to SatIllustrious models. But Experimental model was interesting and generally SatIll has eyes and face quite stable (and repetitive)
Anyway, as a quick attempt to improve the quality is to try SatIllustrious models. They stable enough, but you should do it in elemental mode to avoid washing out your concepts.
IN01, IN02, OUT01, OUT02 blocks - for color
IN08,IN07,OUT08,OUT07 - blocks for finer textures
Thanks to everyone for the comments; they mean a lot to me. I'm taking each and every piece of advice into account. Also, if anyone wants to experiment with the model on their own and has any good results, I'd really appreciate it if you could share the process with me so I can try to replicate it.
Everything they said before, Everything they said before, Homoludens is an expert and there is little he could add extra, good luck with your models
@Sateluco not expert... It still a magic to me.. How to keep fingers and eyes intact when redrawing the dataset ))
im actually testing a refined version, i added a little of a illustrious model from @Sateluco and so far, its working nice... i upload it soon.
And no, i dont sleep. :P
@theNNNNdude I want to ask you one thing. When you integrate some other resources into your model constantly compare the results. Try hard to get some new result on these merges or fixes. To bring any profit to the community the result should have some new useful features.
@homoludens this is for super merge extension, right?
@theNNNNdude In A1111 yes.. I used to use Super Merger extension until it got broken in some functionality. Right now I'm doing elemental merges in ComfyUI only. It is easier or more control.
@theNNNNdude No matter what tool you are using, avoid producing too close model, because there is already too much very similarly looking models on CivitAI.
@homoludens too bad i dont understand cumfyui... i tried, but i get lost on that thing...
@theNNNNdude @belember requested the guide for ComfyUI. If you also neede it let me know what exactly scenario or task you want to be guided with. I will try to find the time for it.
@homoludens i dont even know where to start...
New vIllust-01 seems nice! Not sure if it made any significant changes towards increasing fantasy elements, but the environments are definitely not as plain/boring as v1, but could still be improved. Most of my prompts include fantastical-realism, dreamy magical energies/auras and all kinds of weird, abstract scenarios, but very few checkpoints ever seem to latch onto those ideas (CinEro Illustrious is great in that regard).
Anyway, even with some Illustrious added in, both of your models still feel and perform like Pony to me and imo, that descriptor still seems more appropriate than SDXL for now - though the lines between Illu, Pony and SDXL are certainly blurrier these days, and that's not a bad thing!
@AFD_0 thanks, im actually working on a better version, and yes my goal is to integrate all that surrealism and fantasy into realism, in some way. The first too attemps where clumsy, im trying to refine it block by block now, but its a high learning curve hahaha.