Iscrizione PixAI

SD

Anything V3 FIXER [Quality Fix - Tool and model] (Read description!)

Anything V3 FIXER [Quality Fix - Tool and model] (Read description!) - AI Model cover image
Caricato su
19 lug 2024, 00:52
Utilizzi
5.9k
Recensioni
Eccellente  (2)
Passaggi di campionamento

20

Metodo di campionamento

Euler a

Scala CFG

6

Negativo

(worst quality, low quality:2), (bad quality:1.4), normal quality, (bad-artist-anime:0.9), (by bad-artist:0.9), (EasyNegativeV2:1), (bad hands:1.4), (interlocked fingers, bad eye, unclear eyes, undetailed eyes, missing limbs, fused limbs, broken limbs, fewer digits, extra digits, mutation, deformity:1.3), (bad anatomy:1.1), white background, monochrome, watermark, (worst quality, low quality:1.4), ng_deepnegative_v1_75t, bad-hands-5, verybadimagenegative_v1.3

Permessi
  • Allow image generation & sharing
  • Permettete agli utenti di scaricare il vostro modello
  • Usi commerciali

Descrizione

How to use: Use to enhance from a generated image of Anything V3 with 0,4-0,6 weight and face fix then use it again to enhance it. Follow the full steps:

  1. Generate low resolution with 'QualityFix v1' / Load original image generated with Anything V3. ▪️ Optional: Use LoRA to fix hands.
  2. Select face fix. Fix eyes with 'QualityFix v1' with 0,4-0,6 weight with 'use as a base' option with a low priority. (Cost around 50 credits with 20 steps) ▪️ Don't forget to copy-paste the same prompts, negatives and seed for best results!
  3. Enhance your x768 free image. I recommend to x1.4, with 0,4 weight and 30 Euler a steps (it will cost like 700 credits) up to 0,6 weight and down to 20 steps. Use 'QualityFix v1' again if you want less changes... or with 'Enhancer v1' [COMING up in the future if I find a way to improve it] if you want more changes... or any other model you like that has better hands. ▪️ I recommend these weights: 0,2 (almost no changes) ; 0,3 (minimal changes), 0,4 (standard) and from 0,5 to 0,6 (some changes).

Do you like Anything V3 but you think it is outdated? Would you like to fix your older generations? Well... Here I remade the original Anything v3 with a new way to polish the final results from the original generated images of Anything V3. It should help a little with eyes and details... and very subtle, but it helps with hands too (not a lot, very little). The results should be better than using the original Anything V3 model because it polishes the result. You will still have the same issues but you can use it to fix your old Anything V3 images or to make new ones. This model can also generate images but I also recommend you the original Anything v3 instead.

TL;DR So what this tool does? Basically more contrast, polishing and better lines overall.

Note: Still has the issues of Anything V3 and it only fixes eyes and quality, consider to enhance this model with another one with good hands and use LoRA with hands and give it more steps (30+) if you want full body. One of the worst issues with Any v3 is that making it look was very hard and the eyes are very mixed up. With this tool you'll get better eyes and quality, hands and weird random artifacts might be your only issues.

Comparison for resolutions: (Same seed and with face fix) x1024 https://pixai.art/artwork/1770867912636376104 Any v3 x768 then my full steps https://pixai.art/artwork/1770869449546802546

Commenti1

Artwork made by Anything V3 FIXER [Quality Fix - Tool and model] (Read description!)

1
HomeModelliMio.2Profilo
HomeModelliMio.2Profilo