Models>
ZavyMix
Quick Generate
Model Description by Creator
Come join my Discord!
This model will always remain publicly available for download. It is important to note in this scene that full exclusivity will never be considered. If you have any commercial proposals or are interested in engaging in commercial activities beyond simply selling images created with the model, I encourage you to contact me through Discord. I am open to further discussions and exploring potential collaborations in these areas.Updates
29/07/2023: Inpainting model released based on ZavyMix_v1.
27/07/2023: ZavyMix_v1 released.
27/06/2023: Inpainting model released based on ZavyMix_b1.
24/06/2023: ZavyMix_b1 released.
15/06/2023: Inpainting model released based on ZavyMix_a2. Introduction
The idea behind developing this model stemmed from the frustration experienced with existing models, which often failed to capture enchanting magical effects while maintaining a sense of realism in human subjects. The generated images would either appear flat, with an isolated realistic human in the foreground and a flat magical background, or lean more towards an anime style. Only through extensive prompting effort could somewhat satisfactory results be achieved.
Dissatisfied with these limitations, I made the decision to invest significant effort into developing my own model that addressed this issue and offered a unique solution. The outcome is an exceptionally capable model that excels in both hyper-realism and semi-realism, with an added touch of enchanting beauty that enhances the overall image. With this model the subject is more likely to blend into the scene and become a part of it. Unexpectedly, the model also proved to be highly effective in the realm of macro photography. Its inclination towards magical effects allows for the creation of fantastical insects, flowers, and other subjects at close-up, resulting in captivating compositions. This same effect holds true for bodypainting and alien subjects, as the model maintains a high level of coherence in generating these types of images.
I kindly request that you share your creations both here and on my Discord server, as I would greatly appreciate the opportunity to see them and motivate me to spend more time in further ventures here.
Pros and cons
Pros
The subject is more likely to blend into the scene and become a part of it. It is less likely to be isolated in front of a plain background and more likely to have elements of the scene in front of the subject, even with trained LORA subjects.
Achieve hyper-realistic levels of detail in close-up shots, particularly excellent for macro photography.
Capture and highlight excellent skin texture.
Create enchanting magical effects while maintaining a sense of realism in the subject.
Explore a wide range of artistic approaches, including hyper-realism and 2.5D that goes beyond narrow subject-focused compositions.
It is possible to maintain a realistic or close-to-realism aesthetic even when incorporating more painterly and anime focussed LORAs, up to a certain extent of course. By employing blockweighting techniques on your LORAs, you can further enhance the ability to achieve a realistic or close-to-realism outcome with high coherence to what your LORA is intended to do.
Highly adept at creating alien or bodypaint-inspired designs and concepts.
Very adept at creating NSFW. If that’s not to your liking, mentioning the usual prompt terms in negative prompt will keep it safe.
Both Clip Skip 1 and 2 function effectively. It depends on each individual image, as it varies which option produces more visually appealing results in specific cases.
Cons
The toes and fingers may exhibit a slightly more randomised result than intended, but this outcome is partially expected as a result of merging the subject more closely with the overall environment, instead of separating it clearly from it.
On a simple prompt, the model tends to lean towards generating responses that are more feminine in nature compared to the average.
Roadmap
In-painting model release.Done.Update the model to minimise some of its drawbacks and incorporate a more current blend of models.Done.Models inspired by anime and painterly styles.
Pioneering uncharted LORA subjects (withholding specifics to prevent preemption).
Tips
To better understand the preferences of the model, individuals are encouraged to utilise the provided prompts as a foundation and then customise, modify, or expand upon them according to their desired objectives.
Avoid using face restoration techniques as they are often unnecessary and can lead to inferior results compared to the original model.
Consider utilising Dynamic Thresholding as a method to control CFG. This technique can help you get improved results.
If you find that the details in your work are lacking, consider using adetailer if you’re unable to fix it with prompt alone. Adetailer or similar tools can enhance and enrich the level of detail, resulting in a more compelling output.
For even better LORA results, make use of the LORA block weighting. By incorporating that, you can further enhance the outcome and achieve superior results in the LORA process.
Avoid using underscores in your prompt, unless part of the LORA or TI.
To maintain optimal results and avoid excessive duplication of subjects, limit the generated image size to a maximum of 768x768 pixels or 768x1024 pixels (or vice versa). If you require higher resolutions, it is recommended to utilise the Hires fix, followed by the img2img upscale technique, with particular emphasis on the controlnet tile upscale method. This approach will help you achieve superior results when aiming for higher resolution outputs.
Prompts and TI
Recommended positive prompts: Best quality, masterpiece, absurdres,
Recommended negative prompts: low quality, worst quality, lowres, monochrome, signature, text, logo
You are further encouraged to include additional specific details regarding the desired output. This should involve specifying the preferred style, camera angle, lighting techniques, poses, color schemes, and other relevant factors.
For negative prompts, consider employing Textual Inversion techniques. The following are recommended: easynegative, ng_deepnegative_v1_75t, and one of the various hand TI's like badhandv4. However, there are other TI available that potentially yield more interesting results. Feel free to experiment with different TI and share your findings!
Recommended settings
Vae-ft-mse-840000-ema-pruned.vae.
Euler A, DPM++ 2M Karras, DPM++ SDE Karras.
Steps 20~40 (lower range for DPM, higher range for Euler).
Hires upscaler: UltraMix_Balanced, ESRGAN-4x or 4x_foolhardy_Remacri depending on image.
Hires upscale: Whatever maximum your GPU is capable of, but preferably between 1.5x~2x.
CFG scale 5-8 (unless you use Dynamic Thresholding).
Clip skip 1 or 2 depending on style.
Model recipe
The recipe for the model is quite complex, as it involves utilising different weight settings for each individual input and output layer in each model used in the mixture. Consequently, the following instructions may not yield results that closely resemble this specific model. However, the fundamental recipe is as follows:
Model A:
fantasticmix_v65
Model B:
etherRealMix2
Model C:
profantasy_v22
Model A:
ZavyMix_a1b
Model B:
cyberrealistic_v31
Model C:
majicmixRealistic_v10I would like to express my gratitude to the creators of the models incorporated in this mixture, as well as the creators of the models that came before them.
Social media
You are welcome to join my recently created Discord server, where we can engage in discussions, share our experiences in AI, and showcase the things we’ve made with AI. You are encouraged to join and ask any questions or seek additional tips and tricks related to my models or AI in general. Your participation would be greatly appreciated. Furthermore, I will be releasing early testing versions of my models on Discord. If you are interested in being among the first to try them out, I invite you to join the Discord server.
NOTE: Some of the images have had their contrast edited to enhance details. Results straight from generation may vary.
