Deliberate Guide

Deliberate is the first public Stable Diffusion model that aims to produce everything with minimal prompting and effort. It has enormous dataset and focuses on generating like Midjourney, trying to minimize random results and generating what you describe in the prompt. An older but still powerful model with a beginner-friendly prompt style.

This Model Excels In

• Generating solo male and female subjects, both NSFW and SFW.
• Realistic images from portrait to full-body shots.
• Creating accurate subjects and compositions.
• Crafting various subjects/landscapes.
• Recognizes many art/artist styles.
• Great focus and blur usage.
• It has certain trigger words to improve quality of result.

What To Be Aware of

• This model relies on 3D renders to create images.
• Shorter prompts usually generate better results.
• It tends to create blurry background (bokeh) without prompting.
• Doesn't require terms like extremely detailed, masterpiece, best quality but they can be used. Best option of using masterpiece is not using any negative prompt.
• Does not perform well with generating genitalia.
• Struggle with generating multiple subject images, tends to generate female focused solo scenes.
• cozy trigger word can cause side effects on short prompts. (winter outfits, cold feeling etc.)

Prompting Style

Deliberate prefers shorter prompts and requires negative prompts to generate anatomically accurate, high quality images. User should use deformed, mutated, bad anatomy, ugly like terms with unwanted art styles in negatives to get best results.

Examples for Trigger Words

Deliberate supports certain trigger words that will create images with a certain style. To demonstrate the difference, the table below shows a base image (generated with no trigger words) compared to the same image generated changing only the trigger word used. The trigger words used are as follows:

1. Base image, no trigger words
2. cinematic
3. cozy

Base Image Prompt

portrait of a beautiful woman, front view, {{trigger word}}, outdoors

Recommended Settings

• Steps: 20-30 (22)
• CFG Scale: 6-7
• Sampler: Euler a

Prompt Component Order

1. Medium/Style
2. Subject
3. Angle/View
4. Subject Details
5. Trigger Words
6. Additional Details
7. Quality/Realism Modifiers
8. Lighting/Shadow

Example of Recommended Prompt Structure

portrait of a woman standing in a forest, closeup, beautiful, cinematic, ultra realistic, natural skin, soft light, soft backlighting, soft shadows

Sample Prompts

Beach Scenery

Positive Prompt : portrait of a woman in bikini, sitting on beach, medium breasts, tanlines, (cinematic:1.2), (cozy), amazing beach scenery, (natural skin:1.2), hyperrealistic, highly detailed, 8K, UHD, (depth of field:1.3), [bokeh], soft focus, sunshine, cinematic light

Negative Prompt : deformed, mutated, distorted, disfigured, (ugly, disgusting, doll:1.2), uncanny, bad anatomy, wrong anatomy, (poorly drawn:1.4), (worst quality:1.2), extra limb, missing limb, floating limb, dull, monochrome, blurry

Soft Brush Painting

Positive Prompt : illustration, 2D portrait, nsfw, 1woman, upper body, pursed lips, (large breasts), slim waist, choker, sharp details, sharp linear art, (color toning), [color fading], (soft strokes, soft brush painting)

Negative Prompt : deformed, mutated, distorted, disfigured, (ugly, disgusting:1.2), uncanny, bad anatomy, wrong anatomy, (poorly drawn:1.4), (worst quality:1.2), extra limb, missing limb, floating limb, monochrome, blurry, 3D, ((render, cgi)), asian, japanese

Cosmic Orb

Positive Prompt : digital art, cosmic dimension inside of crystal orb, prismatic colors, refraction, (dispersion of light:1.3), (cinematic quality:1.2), {liminal space|(abyss:1.3)}, amazing background, volumetrics dtx, ray tracing, (cgi), (octane render, unreal engine:1.4), depth of field

Negative Prompt : deformed, mutated, distorted, disfigured, uncanny, (poorly drawn:1.4), (worst quality:1.2), (lowly detailed:1.5), monochrome, blurry, dull composition, cartoon, illustration

Prompt Style Comparison

Below, you can see the difference in generation quality between applying only the general prompt style and applying the recommended style specific to this model.

Final Thoughts

Despite being an older model, Deliberate is still a strong candidate for creating realistic renders and impressive arts with correct prompting. Trigger words are essential when it comes to quality of realistic images and requires certain negatives to generate best possible images, but the learning curve for understanding how to use them is still accessible even to those newer to AI image generation. You'll surely be impressed by what you can create with this classic model.