Update README.md

README.md

@@ -21,38 +21,35 @@ widget:
 Base model: https://huggingface.co/Linaqruf/anything-v3.0.
 Used 'fast-DreamBooth' on Google Colab and 768x768 images for all versions.
 
-I don't recommend using V1, but i've included it anyways.
-
-V1: Trained with around 1300 images (from danbooru), automatically cropped.<br />
-V2: Trained with 400 handpicked and handcropped images, and for the number of steps each model indicates in its name (for example, 'sketchstyle_v2-16000-2500txt_enc.ckpt' is trained for 16k steps, and 2500 steps for the text encoder).<br />
-V3: Trained with the same images as V2, but with 'style training' enabled. Every 10k steps -> 1k steps for the text encoder. <br />
-V4: On training with 407 images, including 'captions' for each image. Not yet avaliable 'officially' (no info on how to use it correctly), trying to reach 100k steps. Will update info later.
-
-Lots of images include visible nipples so, in case you don't want nsfw images, include nsfw/nipples on the negative prompts.<br />
-Also, breasts tend to be large/huge so for anything smaller than that, use medium/small breasts on the prompts and large/huge breasts on the negative prompts, and skin tone tends to be 'tan' because of the same reason,
-so including tan/dark skin on the negative prompts and pale skin on the prompts will help to get whiter tones, though good anatomy gets affected.
-
-Be sure to have this number over 75, and higher if using more steps. Less than that will give bad results, with artifacts.
+**Versions:**
+- V1: Trained with around 1300 images (from danbooru), automatically cropped.
+- V2: Trained with 400 handpicked and handcropped images.
+- V3: Trained with the same images as V2, but with 'style training' enabled.
+- V4: Trained with 407 images, including 'captions' for each image.
+
+**Recommended to use:**
+- V4-42k (pretty good style and anatomy, might be the best)
+- V3-40k (decent style and anatomy)
+- V4-10k (good anatomy, meh style)
+- V4-100k (good style, bad anatomy/hard to use, useful with img2img)
+
+**To have in mind:**
+- Make sure to add nsfw/nipples/huge or large breasts in the negative prompts if you don't want any of those.
+- Skin tone tends to be 'tan', use dark skin/tan on the negative prompts if its the case, and/or pale skin in the prompts.
+- The closer you are to 768x768, the better, but you can get good results in any resolution.
+- Using img2img to change the style of another image generally gives the best results, examples below.
+
+Pay attention to this number. Normally going below 75 generates bad results, specially with models with high steps like V4-100k. Best with 100+
 ![Screenshot_1.png](https://s3.amazonaws.com/moonup/production/uploads/1671505643175-633520c031a2be3938c9f8f5.png)
 
-Token: 'sketchstyle' (depending on the model downloaded, you might have to use it or not, if too much artifacts appear try to not include it, and if the style isn't apparent try to include it and/or strengthen it)<br />
-
-The closer you are to 768x768, the better, but you can get good results in any resolution.
-
-_This model works specially good when changing the style of another image_, so you could generate images with another model that you know works well with anatomy and such, and then use this model to change its style (examples below).<br />
-I recommend, if using V2, to use the 10k steps version as it has the least anatomy errors, and use the 16k steps version when doing img2img.
-
-Tried using this negative prompt instead: 'realistic,3D,3D Game,nsfw,lowres, bad anatomy, bad hands, text, error, missing fingers, bad hands, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts,signature, watermark, username, blurry, artist name, bad face, bad mouth'<br />
-And the results tend to be better, so you may use that instead of what i used for some of the examples.<br />
-Adding this prompt: 'less than 5 fingers, more than 5 fingers,bad hands,bad hand anatomy,missing fingers, extra fingers, mutated hands, disfigured hands, deformed hands' might improve hand quality, try adding parenthesis to strengthen it, 1 or 2 should work fine.<br />
-You might use this for inpainting as it affects the image (and might do it for worse).<br />
+Token: 'sketchstyle' (if used, anatomy may get affected, but it can be useful for models with low steps to get a better style)<br />
 
-Limitations and known errors:<br />
-Might struggle with anatomy, specially hands. Generating lips could be a little tricky but in portraits/close ups it works fine. Eyes also lack detail in general, and might come with artifacts.<br />
-Tends to generate large breasts and tan skin, with skimpy clothes and sometimes with jewelry.<br />
-Sometimes outlines aren't very clear, or it appears there aren't any at all.<br />
-V3, for some reason, adds something to the bottom eyelashes, might be lack of training steps.<br />
-For some of these, doing a little bit of inpainting should work, if it doesn't, try using different versions of the model.
+**Limitations and known errors:**
+- Not very good anatomy
+- Sometimes it generates artifacts, specially on the eyes and lips
+- Tends to generate skimpy clothes, open clothes, cutouts, and similar
+- Might generate unclear outlines
+Try using inpainting to fix these.
 
 
 # Comparison between different versions and models