THIS MODEL IS V-PRED
I tried merge model to give a 3D appearance while keeping the face in an anime style.
enjoy it!
🟦 Recommended Settings
Positive Prompt
masterpiece, best quality, absurdres, highres, anime coloring, 3D
Feel free to add shiny skin, humid skin to the prompt depending on your preferences.
Negative Prompt
lowres, bad quality, low quality, worst quality, text, error, fewer, extra, missing, jpeg artifacts, watermark, unfinished, displeasing, oldest, early, chromatic aberration, signature, artistic username, scan, abstract, 4koma, surreal, abstract, dissolving, multiple views, grey background, sepia, futanari, furry, lips
🔧 Recommended Generation Settings
Sampler:
Euler a,DPM++ 2M SDEScheduler:
SGM UniformCFG Scale:
3–5Resolution Options:
960x12801280x9601128x1128832x12161216x8321024x1024
Sampling Steps:
24–30 stepsHires Settings:
Denoising Strength:
0.3–0.4Upscale Steps:
10–15 steps
Description
I have addressed the issues from the previous version and implemented several improvements.
fixed the issue in v4.0 where faces appeared excessively elongated or had severe style bias.
Enhanced overall image clarity and sharpness.
FAQ
Comments (4)
amazing 3d model
This model seems to need a DPIPE.safetensors VAE, which I haven't found anywhere. Without it the result is basically unusable (too noisy). Any ideas?
this model has a built-in vae, so no extra vae is needed for normal output. If your images look like pure noise, check your v-pred configuration. You must have ztsnr enabled for the model to function properly.
i use this vae
Anime SDXL VAE DPipe Prototype
@zzkz Thank you. It seems this doesn't work well with default AUTOMATIC settings but it does with ComfyUI when you use the built-in vae properly as you said.