AI Content Glossary: Models, Prompts, Credits, C2PA & More Explained
Plain-English definitions of AI content terms: diffusion models, text-to-video, LoRA, ControlNet, voice cloning, C2PA, SynthID, credits, open weights, MCP and more.
Generative AILarge language model (LLM)Diffusion modelText-to-video / image-to-videoPromptNegative promptSeedInpainting & outpaintingUpscalingLoRAControlNetCharacter consistencyLip syncText-to-speech (TTS)Voice cloningSpeech-to-text (ASR)AI dubbingContent Credentials (C2PA)SynthIDWatermarkCommercial use rightsCreditsOpen weightsComfyUI workflowAgent SkillMCP (Model Context Protocol)Brand voiceAI disclosureScaled content abuseAI humanizer
- Generative AI
- AI models that create new text, images, video, audio or code from a prompt, rather than only classifying or predicting. Everything in this directory is built on them.
- Large language model (LLM)
- A model trained on huge amounts of text that predicts the next word. GPT, Claude, Gemini, Llama and Mistral are LLMs; writing assistants and chatbots run on top of them.
- Diffusion model
- The main technique behind image and video generators: the model learns to turn random noise into an image step by step, guided by your prompt. Stable Diffusion and FLUX are diffusion models.
- Text-to-video / image-to-video
- Generating a video clip from a written prompt, or animating a still image you supply. Image-to-video gives much more control over how the shot looks.
- Prompt
- The instruction you give a model. For images and video it describes subject, style, camera and lighting; for text it sets the task, audience, tone and constraints.
- Negative prompt
- Things you tell an image model to avoid ("blurry, extra fingers, text"). Supported by Stable Diffusion-style tools; many newer models ignore it.
- Seed
- The random number a generation starts from. Reusing the same seed and prompt reproduces an image, which makes small, controlled changes possible.
- Inpainting & outpainting
- Regenerating only a selected part of an image (inpainting), or extending the canvas beyond its edges (outpainting). Adobe calls these Generative Fill and Generative Expand.
- Upscaling
- Increasing an image or video's resolution with AI that invents plausible detail. Useful before print or 4K export; it can also smooth away texture.
- LoRA
- A small add-on file that teaches an open image or video model a style, product or character without retraining the whole model. Shared widely for FLUX, Stable Diffusion and Wan.
- ControlNet
- A way to guide an image model with a pose, depth map, edge drawing or sketch, so the output follows a layout you choose.
- Character consistency
- Keeping the same face, outfit or product across many images or video shots. Tools solve it with reference images, "characters" features or LoRAs.
- Lip sync
- Matching a face's mouth movements to an audio track. Used by avatar tools, dubbing apps and some video models.
- Text-to-speech (TTS)
- Turning written text into spoken audio. Modern TTS controls emotion, pacing and pronunciation and can speak dozens of languages.
- Voice cloning
- Creating a synthetic copy of a real voice from a recording. Reputable tools require proof of consent from the voice owner.
- Speech-to-text (ASR)
- Automatic transcription of spoken audio into text; the basis of subtitles, show notes and text-based video editing. Whisper is the best-known open model.
- AI dubbing
- Translating the speech in a video into another language and re-voicing it, often in the original speaker's cloned voice and with lip sync.
- Content Credentials (C2PA)
- An open standard for attaching tamper-evident provenance data to a file: who made it, with which tools, and whether AI was used. Adobe, OpenAI, Google and camera makers support it.
- SynthID
- Google DeepMind's invisible watermark embedded in images, video, audio and text generated by Google's models, detectable with Google's tools.
- Watermark
- A visible logo or an invisible signal added to generated media. Many free plans add a visible watermark; paid plans usually remove it.
- Commercial use rights
- Permission to use what you generate in work you sell or monetise (client work, ads, monetised channels). Often only granted on paid plans; always check the tool's terms.
- Credits
- The unit most AI media apps bill in. A plan includes a monthly pool; each generation costs credits depending on model, length and resolution.
- Open weights
- A model whose trained weights you can download and run yourself (Wan, FLUX [dev], Stable Diffusion, Whisper). The licence still decides whether commercial use is allowed.
- ComfyUI workflow
- A node graph that chains models and processing steps in ComfyUI, saved as JSON (or embedded in a PNG) so anyone can load and rerun it.
- Agent Skill
- A folder with a SKILL.md file of instructions (and optional scripts or templates) that an AI assistant loads only when a task needs it, such as your brand voice or a report format.
- MCP (Model Context Protocol)
- An open protocol that lets AI assistants call external tools. An MCP server for ElevenLabs, WordPress or Canva lets Claude or ChatGPT generate audio, publish a post or edit a design.
- Brand voice
- A saved description (and examples) of how a brand writes: tone, vocabulary, do and don't lists. Writing tools apply it so drafts sound like you.
- AI disclosure
- Telling your audience or a platform that content was made or altered with AI. YouTube, TikTok and Meta ask for it on realistic synthetic media, and the EU AI Act adds transparency duties.
- Scaled content abuse
- Google's spam policy against mass-producing pages mainly to manipulate rankings, however they were made. AI-assisted content is fine when it is genuinely helpful.
- AI humanizer
- Tools that rewrite AI text to evade AI detectors. We do not list them: they target deception, and detectors are unreliable anyway.