Reference

Glossary

Name: BeModel
Availability: InStock
Rating: 4.7 (6 reviews)

Plain-language definitions of every domain term used across BeModel — for users, journalists, and anyone trying to understand AI photo generation in 2026.

Active avatar: The avatar currently used for new photo generations on your account. Each user has one active avatar at a time but can switch between any avatars they own.
AI avatar: A digital character generated and remembered by an AI model. Unlike a one-off AI image, an avatar persists across generations so the same face and body appear in every photo created with it. BeModel is built around persistent AI avatars.
AI Builder: BeModel's visual editor for designing an avatar from scratch — pick face, hair, body, skin attributes through a UI instead of uploading reference photos. Output is a custom avatar that you own.
Aspect ratio: The width-to-height proportion of a generated photo. BeModel templates render in 9:16 (vertical, optimised for Instagram/TikTok feeds and Stories) or 4:5 (Instagram-friendly portrait).
Credit: The unit of currency inside BeModel. 1 credit = 1 generated photo. 5 credits = 1 generated 5-second video. Plan credits reset each billing cycle; top-up credits you buy separately never expire.
Custom avatar: An avatar built from your reference photos or via the AI Builder. Private to your account, unique to you. Costs 5 credits to create.
Free trial: BeModel's no-card sign-up offer: every new account gets a starter avatar plus 3 free photo generations. No payment method required to claim them.
Generation: The process of producing a single photo or video. Pick a template, click generate; the AI renders your active avatar inside that scene. Photos cost 1 credit. 5-second videos cost 5 credits (via the image-to-video pipeline).
Hyper-realistic: Photo output that's indistinguishable from real photography at typical viewing sizes. BeModel uses the Nano Banana Pro model to hit this quality bar consistently.
Identity consistency: The property of an AI media system where the same character (face, body, hair, skin) appears reliably across every generation — both photos AND videos. The hardest problem in AI image and video generation, and BeModel's main differentiator versus prompt-based tools like Midjourney, Pika and Runway.
LoRA: Low-Rank Adaptation. A technique for fine-tuning a diffusion model on a small dataset (e.g. photos of one specific person) to make the model reliably reproduce that subject. BeModel hides LoRA-style training entirely so users never have to set one up themselves.
Kling v2.6 Pro: The AI model that powers BeModel's video generation, via the kwaivgi/kling-v2.6-pro/image-to-video endpoint on Wavespeed. Takes a starting frame and a motion prompt and renders a 5-second clip that inherits the aspect ratio and identity of the input image.
Image-to-video: BeModel's two-step pipeline for video generation: (1) qwen-image-2.0-pro generates a starting frame from your avatar's reference photos and the template's photo prompt, (2) Kling v2.6 Pro animates that frame into a 5-second video using the template's video prompt. The starting-frame step is what locks identity into the video.
Nano Banana Pro: The AI model that powers BeModel's photo generation — Google's Qwen 2.0 Pro family. Currently one of the most photoreal text-to-image models available, with strong identity preservation under reference conditioning. Also generates the starting frame for video templates.
Pending generation: A photo or video whose generation has been requested but isn't finished yet. Appears in your gallery with a water-fill animation while it processes. The slot becomes the real media automatically when ready.
Plan: A monthly or annual subscription tier. BeModel has four: Starter (100 credits/month), Pro (300), Creator (600), Diamond (1000). All plans include credit refund protection and access to every template.
Preset avatar: A pre-built starter avatar auto-assigned to every new account at sign-up, so the product works the moment you register. Comes paired with 3 free photo generations.
Reference photos: The 3 photos you upload (face, torso, full-body) when building a custom avatar from real-world images. Used to train the avatar's identity; never used to train BeModel's underlying model.
Refund protection: Automatic credit refund when a generated photo is clearly broken (deformed face, doesn't match template, off-avatar). Up to 10% of monthly generations qualify for refund — trigger it by clicking Report on any bad photo in your gallery.
Stable Diffusion: An open-source AI image generation model that powers many tools in the space. Requires prompt engineering, model tuning and a GPU; powerful but technical. BeModel is built on a different model family (Qwen) for better identity consistency.
Subscription: An auto-renewing plan that gives you a monthly or annual allowance of credits, plus refund protection and full template access. Cancel any time — access continues until the end of the current billing period.
Starting frame: The intermediate still image generated by Nano Banana Pro at the start of a video pipeline. Built from the avatar's reference photos and the template's photo prompt, then handed to Kling v2.6 Pro as the visual anchor for the 5-second clip. Locks the avatar's identity into the video from frame one.
Template: A pre-built scene definition that the AI uses as the context for a generation: setting, pose, wardrobe, lighting and (for video templates) motion. BeModel has 500+ templates across casual, hot, selfie, outdoor and other categories. Each template is either a photo template (1 credit) or a video template (5 credits, marked with a VIDEO badge). New templates added every week.
Top-up: A one-time purchase of extra credits added to your account without changing your subscription. Top-up credits never expire. Available in packs of 100, 300, 600 and 1000.

Missing a term you'd like defined? Email us at hello@bemodel.ai and we'll add it.