Sign In

PhoenixMix

13
161
0
Verified:
SafeTensor
Type
Checkpoint Merge
Stats
161
Reviews
Published
Aug 3, 2023
Base Model
SD 1.5
Training
Steps: 35
Usage Tips
Clip Skip: 1
Hash
AutoV2
436DF278BB

Color coded in red is highly recommended

Color coded in yellow is clearly unsure

Color coded in lime green is optional

Sorry I’m late; I was a bit too lazy/nervous to upload if.

Description

So, Basically, this model is a checkpoint merge of the following models:

If these links don’t work, please let me know

All models above belongs to their respective owners/creators, not me (I'm the one crediting them)

Tips

As I said [due to my disability and my low-end device], these are the tips I could give you so far:

  • I don’t know what VAE to use, so please let me know in the comments

  • If your image looks weird or abnormal than what you expected, then you may need to add these to the prompt: i.e. ((best quality)), ((masterpiece)), (detailed) from Rev's; ((8k resolution)), ultradetailed, (perfect quality), highest quality, ((perfect proportion)); fine details.

  • For the negative prompt: i.e. ((worst quality, blurry, watermark, noise, noisy, pixelated, corrupted, glitched, too much contrast, now enough contrast, imbalanced quality)).

  • If your character looks ugly and have messy body parts, you may need to put the prompts like: (((perfect proportion, perfect anatomy, perfect face, perfect eyes, perfect nose, perfect mouth, perfect hands, all 5 fingers, perfect feet, all 5 toes, even body parts))). Unfortunately, I do not know the order/strength so you may need to figure it out :(

  • The negative prompt is basically an opposite of what is above (the third point in tips): i.e. (((worst proportion, worst anatomy, worst face, worst eyes, worst nose, worst mouth, worst hands, missing fingers, more than 5 fingers, worst feet, missing toes, uneven body parts)))

  • The best negative textual inversion to use is <ng_deepnegative_v1_75t> (https://civitai.com/models/4629/deep-negative-v1x) I couldn’t convert it to link for some reason

  • Use the line break if they offer one

  • The best image size depends

  • If your device runs perfectly or can handle things "without burning", then these are following upscalers you can use: Remacri, Universal Upscaler V2 Sharp, Real-Esrgan X2+ (for low ends who wants the best quality)

  • Depending on your device, use steps ranged from 25-45

  • Text guidance should be from 5-7

  • Take the lowest clip skip as possible

  • request me some tips to use this model so I can change them asap

  • Also, sorry for me grammar/spelling, especially when writing on a phone (oof, uncomfortable)

Features

Pros

  • The text input has freedom/unrestricted, making it pretty flexible

  • Most of the time, the model creates well-made quality so I don’t want to get further into this.

  • They did a decent job on poses

  • Easy to generate women and non-binary folks (what I meant is androgynous).

Mehs

  • Loras seem to work but certain prompts, especially related to centaurs, Have a few change to work. Maybe adding enough strength works

Cons

  • Even with negative prompts, the hands and faces seemed off

  • Doesn’t really work great with low ends

  • Inpainting/tiles/refining is difficult to do

  • It can get sexist because when you prompts (such as "man wearing a dress and skirt"), it generates a woman instead of a man. If you want a man instead, then you should strengthen it and put woman/women/female/girls in the negative prompt with enough strength (what I’m saying is by using parentheses).

Origins/Fun Facts

  • This model was originally used to make an unreleased model with the longest name ever, but my phone couldn’t handle it

  • I named this model [PhoenixMix] because Phoenixes are my favorite mythical birds/creatures (besides mermaids). My apology for the bias.

  • This is the first uploaded model I created

Disclaimer

  • if you want to merge this model, then you should definitely credit me. Every time I see merged models/or used ones, only a few credit someone for using it. =(

  • If you generate an image, please provide everything (such as positive/negative prompts, steps, cfg, and models) so I can gain more influence.

  • Do not sell anything (i.e. models and images) that revolves around PhoenixMix, because it fan cause conflicts with an NFT (very controversial) and I don’t like to get into conflicts with anything like those.

  • Again, sorry about the spelling/grammar (although my english seems to be great)

  • Again, comment me if you suggest me