Glossary
Short definitions for the terms the concept pages use: latents, encoders, denoising, guidance, and the parts of a generation pipeline.
Learn
How generation works, from the mechanisms underneath to what each parameter changes in the result.
Latent space
The compressed space a model actually works in, and what that compression decides about every result.
Latent volume
How a video model holds time as another axis of what it denoises, and what that decides about every clip.
Denoising
The loop that turns noise into an output, one question at a time, and what each step really decides.
Text encoders
The network that reads your prompt before the generator sees it, and what it decides about how you write.
Text to image
The foundational pipeline: generate images from text prompts using diffusion models, with control over every step of the process.
Image to image
Transform existing images using a source image as a starting point, controlling how much of the original to preserve.
Inpainting
Selectively edit specific regions of an image using masks, replacing or regenerating content while preserving the rest.
Outpainting
Extend images beyond their original borders in any direction while maintaining seamless continuity.
Image to video
How a still image becomes the first frame of a generated clip, what the frame decides, and what the prompt is left to decide.
Prompts
How positive and negative prompts steer generation, what each modality expects you to describe, and the techniques that hold across models.
Dimensions
How width and height shape composition, which outputs accept them, and the architecture limits that decide safe values.
Seed
Pin the random noise to reproduce identical outputs or create controlled variations from the same starting point.
Steps
How the steps parameter controls the iterative refinement process, with architecture-specific ranges and speed vs quality tradeoffs.
Schedulers
The denoising algorithms that guide diffusion models from noise to image, each with different speed and quality tradeoffs.
CFG Scale
Controls how strictly the model follows your prompt, balancing creativity with adherence.
CLIP Skip
Adjusts which text encoder layer interprets your prompt, shifting between literal and abstract output.
VAE
The decoder that converts the model's latent representation into the final image, affecting color fidelity and detail.
ControlNet
Provides structural control over generation using conditioning inputs like edge maps, depth maps, and pose detection.
LoRAs
Lightweight model adapters that add specific styles, subjects, or concepts without retraining the base model.
IP Adapters
Use reference images to guide generation, enabling style transfer and visual consistency across outputs.
Embeddings
Encode custom visual concepts, styles, or subjects into specialized tokens for consistent generation.