AI Models & Generation Disclosure

Video generation

The video generator uses ByteDance Seedance 2 Mini, with the model identifier bytedance/seedance-2-mini, through Kie. Kie is the API service provider; Seedance is the underlying generation model. Scene and reference-video modes use this same model. There is no hidden video model switcher.

Image generation and editing

The image studio offers these specific model choices:

• OpenAI GPT Image 2 — selector ID gpt-image-2.
• Google Nano Banana Pro (Gemini) — selector ID nano-banana-pro.
• Google Nano Banana 2 (Gemini) — selector ID nano-banana-2, the default selection.

The selected model is shown beside the prompt before submission. All three image models and their editing workflows are served through Kie. The image assistant uses DeepSeek V4.1 Flash, with low reasoning effort, to interpret requests and invoke the selected image model. This conversation model does not replace the image model shown in the selector.

Checking image generation availability…

Image credit prices

Each generated or edited image is charged by the selected model and resolution:

• Nano Banana 2 — 1K: 3 credits; 2K: 4 credits; 4K: 5 credits.
• GPT Image 2 — 1K: 4 credits; 2K: 5 credits; 4K: 6 credits.
• Nano Banana Pro — 1K: 5 credits; 2K: 6 credits; 4K: 7 credits.

Auto uses 1K and the 1K price. The studio displays the price before submission. Generation and editing have the same rates, and image and video generation share your credit balance. Failed image tasks return their deducted credits.

What a result means

Images, movement, scenes, dialogue and voices produced by the generators are synthetic. Outputs may differ from prompts or references and can contain visual or factual errors. They are not recordings of real events. Model names identify the technology; they do not imply affiliation, endorsement or ownership of those brands.

All model choices follow our Content standards. Use the reporting channel for unsafe results, inaccurate model labels or concerns about your rights. Read the Terms of Service before creating or sharing content.

Content review

Content review uses DeepSeek V4.1 Flash (deepseek-v4-1-flash) through Kie, with reasoning effort low. If this service times out or is unavailable, GPT-6 Luna (gpt-6-luna) through Kie applies the same rules. A rejection or a request for human review never triggers this fallback. Generation success alone does not authorize delivery. Video frames are sampled at four frames per second using Cloudflare Browser Run; actual audio is transcribed with Whisper large-v3-turbo (@cf/openai/whisper-large-v3-turbo) on Cloudflare Workers AI. Real-person creation is paused. See Content standards for release controls and limitations.