AnimationDiff Guide

This model is currently being beta tested and subject to change.

AnimationDiff combines Realistic Vision's photorealism with animation, creating unique gif-like images. This guide offers a detailed exploration of its features and nuances.

This Model Excels At

  • Creates outstanding realistic images
  • Can create hentai and cartoon images
  • Creating high quality starting images with short prompts and weights
  • Using Particle effects and lighting

What to be Aware of With This Model

  • Operates using ComfyUI
  • The height and weight parameters solely affect the rendered image, not the dimensions of the final animation.
  • Configured to ping pong mode, reducing animation frames from 16 to 8
  • Creates a static image before animating
  • Lacks motion controls
  • Utilizes special operators
  • Lacks keyframe prompting
  • Works with multiple subjects and POV shots when prompts adhere to the guidance in Prompting 101.
  • Steps: 25 - 30
  • CFG Scale: 3 - 7
  • Sampler: Euler /DPM++ 2M Karras

Prompting Style

AnimationDiff uses the general style prompting, and performs best with small prompts. This implementation of the model does not support controls for camera movement.

Prompt Component Order

  1. Subject
  2. Details
  3. Small Details
  4. Quality

Sample Prompt Order Example:

a woman wizard casting an ice spell,gorgeous,floating ice crystals,(best quality)

Additional Comfy UI Syntax

While most syntax in Comfy UI is similar to Automatic1111, there is one major difference in the way that they two styles handle nested weighting. For example, in ComfyUI, ([term:0.2]:1.3) makes the weight term .26 ,while in Automatic1111 it's 1.1 as its subtracts 0.2 instead of multiplying.

Sample Prompts

Ice Magic Dancer


Positive Prompt : (a gorgeous polish woman:1.5),(windy dancers, [snow veil:headgear]), ice crystals, illuminating particles, (shallow focus:1.5), cinematic shot, dramatic nighttime lighting, deep shadows,(best quality:1.5), (convey metamorphosis:1.2)

Negative Prompt : haphazard, deformed,teeth,(3d render:1.5), low quality, low detail,(worst quality:2), (hands:1.3), dull composition, static components

Soccer Balls


Positive Prompt : pov of a window,[soccer ball enters:travels to end of frame:0.9][soccer ball vanishes:0.3], [soccer ball enter from left:ball horizontal motion:0.9][soccer ball vanishes:0.3], (symmetrical motion), animated perfect loop, (best quality:2)

Negative Prompt: (3d render:1.5), 3d motion, (worst quality:2)

Male Man Dancing


Positive Prompt : a latino muscleman short fade hairstyle, wearing male boxerbriefs, (impossible underwear, tight underwear),([shallow depth of field|deep depth of field]:1.3), thick body, [mustache:goatee:0.9],(gyrating hip thrusts), (closeup), (ultra quality)

Negative Prompt : (long hairstyles:1.5),(3d render:1.5), low quality, dull composition, single camera, worst quality, normal quality,ugly, deformed, disfigured

Prompt Style Comparison

Below you can see the difference in generation quality between applying only the general prompt style in Prompting 101 and applying the recommended style specific to this model.

general.webpcloseup portrait shot, a woman standing in the forest, masterpiece, extremely detailed, soft light, soft shadows, soft backlighting, (best image, best quality:1.5)

Final Thoughts

AnimationDiff, currently in beta, seamlessly combines photorealistic quality with animation. It operates on ComfyUI instead of automatic1111 and has ping pong mode active which limits the animation frames to 8 instead of 16. Despite this, AnimationDiff excels in creating realistic and cartoon images. With strategic prompts, users can unlock its potential for intricate animations. As development progresses, further enhancements are anticipated.