SDXL

NoobAI VPred+IterComp (test)

NoobAI VPred+IterComp (test) - AI Model cover image
Diunggah pada
7 Nov 2024, 08.45
Penggunaan
5.7k
Ulasan
Bagus  (3)
Langkah sampling

20

CFG scale

6

Negatif

blurry, distorted, film grain, grainy, yellow theme, error, worst quality, sketch, signature, monochrome, jpeg artifacts, scan artifacts, compression artifacts, artist name, artist logo, patreon username, watermark, (bad hands:1.3), deformed, (extra finger:1.3), steam, steaming body, sad, frown

Izin
  • Izinkan pembuatan & berbagi gambar
  • Izinkan pengguna mengunduh model Anda
  • Penggunaan komersial

LoRA yang Direkomendasikan

Deskripsi

This is a stabilization of the v-prediction test of NoobAI. Weights from IterComp are added onto the model using sd_mecha with perpendicular component add difference. (this prevents noise schedule from being affected). Then the original clip is merged into the result.V-prediction support should be in the next version of Automatic1111. In ComfyUI, use the ModelSamplingDiscrete node. In SwarmUI, select "Vpred ZSNR" under "Prediction Type" in "Edit Metadata" on the model.Iterative Composition-Aware Feedback Learning: https://arxiv.org/pdf/2410.07171APG scaling helps with dynamic range.

Komentar1

Karya oleh NoobAI VPred+IterComp (test)