RAP: A Simple Merge for Realism

RAP, is a simple merge of three distinct checkpoints—Realistic Vision, UniverseStable (Analog Photo), and PornMaster-Amateur. The core objectives that guided this merger:

1. Balanced Anatomy Tokens:

   • Produce captivating images of both human and non-human subjects while maintaining a delicate balance. It avoids an overabundance of negative anatomy tokens, ensuring a harmonious visual experience.

2. Subtle Body Enhancements:

   • Leverage the ‘Hires fix’ feature up to 0.4 to achieves realism without veering into hyperbole. No washboard abs or exaggerated ribcages—authentic representations.

3. Anime-Free Accuracy:

   • When paired with LoRAs, capture accurate facial and body features. Prioritize authenticity without anime-style distortions.

4. Diverse Ethnic and Age Groups:

   • Through simple prompting, generate wide ethnicities and age ranges.

* Please reference additional images under my profile, the majority of them use this checkpoint, but I uploaded the images prior to publishing the checkpoint so it isn't listed as a resource. It is included by name in the generation information.

Recommended Tools:

• Key Tools: ‘Hires. fix’ and ‘ADetailer’

  • Enable the following two options in A1111

    • Hires fix: show hires checkpoint and sampler selection 

      (requires Reload UI)

    • Hires fix: show hires prompt and negative prompt 

      (requires Reload UI)

  • Change the following option in 'ADetailer'

    • Max models: 5

• Face Restoration: While not the norm, consider using it only for small faces (e.g., full-body images). Otherwise, rely on the specialized face, eyes, and lips models within ‘ADetailer’.

• Negative Embeddings (use only one):

  • RAP Negative Embedding - v1.0 | Stable Diffusion Embedding | Civitai

  • 0001SoftRealistic Negative and Positive Embeddings - v8 | Stable Diffusion Embedding | Civitai

Workflow Template:

Feel free to tailor this workflow to your preferred UI.

Prompt:

[professional::0.3][intricate:0.3:0.7][digital:0.7]-photography, {prompt}, picture-perfect

Negative Prompt:

RAP.negative.emb

Negative Prompt (Alternative):

(cartoon, anime, render, rendered, 3d render, 3d cgi, waifu, airbrushed, oil painting), conjoined, mutated, twins, sisters, multiple, (text in corner:1.9), 0001SoftRealisticNegativeV8-neg

Sampler: DPM++ 2M Karras, DPM++ SDE Karras

Sampling steps: 24

CFG Scale: 6

Clip Skip: 1-2

‘Hires. fix’ Settings:

• Upscaler: 8x_NMKD-Superscale_150000_G, 4x-UltraSharp, WaifuGAN_v3_30000, DAT_x4, or 4x_NMKD-Siax_200k

• Hires sampling method: DPM++ SDE Karras, DPM++ 2M Karras

• Hires steps: 20

• Denoising strength: 0.25-0.4 (higher denoise can help with hands)

Optional ADetailer Settings (only non-defaults):

Recommended

• ADetailer model: face_yolov8n.pt, ADetailer denoising strength: 0.2-0.4, ADetailer use separate steps: True, ADetailer steps: 20 (* Don’t input a prompt, allow ADetailer to reuse the main prompt)

• ADetailer model 2nd: mediapipe_face_mesh_eyes_only, ADetailer denoising strength 2nd: 0.4, ADetailer use separate steps 2nd: True, ADetailer steps 2nd: 20 (* Don’t input a prompt, allow ADetailer to reuse the main prompt)

• ADetailer model 3rd: lips_v1.pt, ADetailer confidence 3nd: 0.6, ADetailer Mask max area ratio 3rd: 0.05, ADetailer denoising strength: 0.25, ADetailer use separate steps 3rd: True, ADetailer steps 3rd: 20, ADetailer use separate sampler 3rd: True, ADetailer sampler 3rd: Euler a (* Don’t input a prompt, allow ADetailer to reuse the main prompt)

If needed

• ADetailer model 4th: hand_yolov8s.pt, ADetailer prompt 4th: "close-up hand, wrist, model, detailed skin texture, (blush:0.5), (goosebumps:0.5), subsurface scattering", ADetailer denoising strength 4th: 0.2-0.3, ADetailer use separate steps 4th: True, ADetailer steps 4th: 20, ADetailer use separate sampler 4th: True, ADetailer sampler 4th: DPM++ 2M SDE Karras

• ADetailer model 5th: foot_yolov8x_v2.pt, ADetailer prompt 5th: "close-up foot, model, symmetrical, ankle, detailed skin texture, (blush:0.5), (goosebumps:0.5), subsurface scattering", ADetailer denoising strength 5th: 0.2-0.3, ADetailer use separate steps 5th: True, ADetailer steps 5th: 20, ADetailer use separate sampler 5th: True, ADetailer sampler 5th: DPM++ 2M SDE Karras

Thanks to the creators of these models for their work. Without them it would not have been possible to create this model.

Realistic Vision V6.0 B1 - V5.1 (VAE) | Stable Diffusion Checkpoint | Civitai

UniverseStable - v7.0(Analog Photo) | Stable Diffusion Checkpoint | Civitai

PornMaster-Amateur - FULL-V6-fix | Stable Diffusion Checkpoint | Civitai

For those seeking to produce NSFW images RAP inherits many of PornMaster-Amateur's prompt enhancements though to a lesser degree. I recommend reading of the checkpoint's documentation.

描述:

v1.0 - Initial Merge

训练词语: