Glossary
21 AI generation terms,
plainly defined.
If a phrase on hilens.ai is jargon (CFG scale, LoRA, inpainting, diffusion), it lives here. Each entry: a one-paragraph definition, an extended explanation, and where relevant a concrete example using hilens models and credit costs.
Credit
aka AI credit, generation credit
On hilens.ai, a credit is the unit of payment for one increment of generation work. Models cost different numbers of credits per generation; credits are purchased in packs and never expire.
Credit pack
aka AI credit pack
A bundle of credits purchased one-time at a fixed dollar price. Larger packs offer a lower per-credit rate. Credits in a pack do not expire and are not auto-renewed.
Pay per generation
aka pay per use, pay per credit
A pricing model where you pay only when you click Generate, debited from a prepaid credit balance. Contrasts with subscription, where you pay a fixed monthly fee whether or not you generate.
Subscription credit
A credit included in a monthly subscription quota that expires when the next billing cycle starts. Distinct from a pay-per-generation credit, which never expires.
Prompt
aka AI prompt, text prompt
The text instruction given to an AI model that describes what to generate. For image and video models, prompts mix subject description, style direction, camera or composition hints, and optional negative prompts.
Negative prompt
Text describing what should not appear in the generated image or video. Used to suppress common failure modes like extra fingers, watermarks, blurry textures, or specific styles you want to avoid.
Reference image
aka ref image, image-to-image input
An image uploaded alongside the prompt to condition the generation. The model preserves visual properties — composition, color palette, subject likeness, framing — from the reference in the output.
Aspect ratio
aka AR
The width-to-height ratio of a generated image or video. Common ratios: 1:1 (square), 9:16 (vertical / TikTok / Reels), 16:9 (horizontal / YouTube), 4:5 (Instagram portrait), 3:4 (editorial vertical).
Seed
A numerical value that initializes the random-number generator inside a model. Fixing the seed makes outputs reproducible — the same prompt + same seed yields the same output.
CFG scale
aka guidance scale, classifier-free guidance
A parameter controlling how closely the model follows the text prompt. Higher CFG = more literal interpretation; lower CFG = more creative freedom. Typical range 4-12.
Inference steps
aka sampling steps, denoising steps
The number of iterative refinement passes the model takes to produce an output. More steps generally means higher quality, with diminishing returns above a model-specific threshold.
Diffusion model
A class of AI generative models that learn to reverse a noising process — starting from pure noise and progressively denoising into an image or video matching the prompt.
Text-to-image
aka T2I, txt2img
A generation mode where the input is text (the prompt) and the output is an image. The most common AI image generation mode.
Image-to-video
aka I2V
A generation mode where the input is a still image and the output is a video animating that image. Often paired with a text prompt describing the desired motion.
Video generation
aka AI video, video synthesis
AI-driven creation of video clips from text prompts, reference images, or both. Output is typically a short clip (5-10 seconds) at fixed resolution.
Identity preservation
aka character consistency, identity lock
A model's ability to keep a person's face or a character's appearance consistent across multiple generations from the same reference. Critical for multi-panel collages and character-driven content.
In-image text
aka text rendering, in-frame text
Text that appears inside the generated image as part of the scene — signage, packaging copy, quote cards, infographic labels. Historically a weakness of diffusion models; in 2026, only the strongest models render it legibly.
Output resolution
The pixel dimensions of a generated image or video. Common values: 1024×1024 (default for most image models), 4K (Nano Banana 2 portrait), 1080p (most video models).
Upscaling
aka AI upscale, super-resolution
Increasing the resolution of an existing image or video using an AI model. Adds detail by inferring what the higher-resolution version would look like, rather than just interpolating pixels.
Inpainting
aka AI inpaint, region edit
Replacing or editing a specific region of an image while keeping the rest unchanged. The user masks the region; the model generates new content matching the surrounding context.
LoRA
aka Low-Rank Adaptation, LoRA model
A lightweight model fine-tuning technique that adjusts a base model to a specific style, character, or subject without retraining the whole model. LoRAs are small (typically 50-200MB) and stack on top of a base.