Do not download if your workflow relies primarily on LoRAs.
If you appreciate the model, leave a thumbs-up. Models with consistently poor appreciation-to-download ratios are removed to keep the space focused. You decide in which direction the development goes.
Highly prompt-sensitive models intended for specialized creative use cases rather than general-purpose generation. Proper prompting is essential for good results.
A low CFG is strongly recommended. Negative prompts and embeddings are usually less important, as crucial parts of that behavior are already merged into the models.
This package contains several distinct models with different intended applications rather than direct variations of one another. The goal is to explore different visual capabilities, aesthetics, and workflows. Community feedback helps determine what works reliably and what still needs refinement.
Nothing groundbreaking, and probably far too late to the SD party. SD1.5 is slowly fading out, so this may become obsolete sooner rather than later.
Description
Reworked model - previously released but not well crafted. Here is a refined version with a better understanding.
FAQ
Comments (7)
Have you tested this model with img2img?
I cant understand why but the output image is always black. I tried different denoise amounts. I thought the new update might have broke it so I tried different models,. nope, always black.
I havent dl your update to the model, im still using aiREalistic_bAISics
Next Ill dl that and see if it works. If It does Ill delete this.
regular generation is great. I like this model for landscapes.
UPDATE: I downloaded your updated version of this model, aiREalistic_pAIXMaster.
Same img2img only outputs blacked out images.
I thought all models do img2img but the output may or may not produce quality results due to what the model was trained on.
Newest update: I said If I were to get img2img working that I would delete this comment post.
I did get it working. But the solution is so odd so I'm leaving this as a note. Its also a chance for me to learn something. with your help that is.
The solution was just a guess. after trying many different things with no change I had the idea of trying a different VAE using the VAE loader. It worked!
Using the vae-ft-mse-840000-ema-pruned. In the most basic workflow for SD1.5 all the VAE connections come from the model loader. The VAE connector on the load image that goes into a VAE Encoder is the only one using the LoadVAE node. IMG2IMG works this way but not when the VAE Encoder uses the LoadCheckpoint Node's VAE.
Strange, I would think that means you're using a custom VAE, one not "configured" to work on IMG 2IMG. I guess? I dont know enough about VAE's to understand why VAE's would work this way. What do you think, @_CreAI_ ?
Thank you greenmonkey3 for this investigation that helps trying to understand this problem. It persists on my side.
Not just with IMG2IMG but with upscalers too. Following Your comment I realize that this is a model-merge problem and not my SDA1111 installation. Thank you for that - as it limits the usability of my models unintentionally by an important factor. I still don't know how to resolve this as I have earlyer merges from August/September last year that does not have this problem while changing the VAE does not solve it for me.
@greenmonkey3 Well - surely you know more about the contextual behaviour behind the UI of A1111 than I do. I don't use VAE, no embeddings, (nearly) no negative prompts, some Loras for fun and to show their positive impact but upscaling anf img2img does not work. With your analyze I will try to figure the last point out and delete this models and change them for functional versions. Thank you for that again!
@greenmonkey3 As long as I merge one of these models in any other model capable of img2img (weight 0.5) it cracks the base for that merge and discontinues being usable for img2img.
Well, actually the problem is solved and the output images are normal.
@_CreAi_ I see you updated the model. Thank you.
AI-r-ealistc is a great environment model. It does humans pretty dang good too.
Looks like we don't have an active mirror for this file right now.
CivArchive is a community-maintained index — we catalog mirrors that volunteers upload to HuggingFace, torrents, and other public hosts. Looks like no one has uploaded a copy of this file yet.
Some files do get recovered over time through contributions. If you're looking for this one, feel free to ask in Discord, or help preserve it if you have a copy.