AI Model Specifications

ByteDance Seedance Lite (V1 Lite) is a budget-friendly text-to-video and image-to-video model available at 480p, 720p, and 1080p in 5s or 10s clips, with six aspect ratios (16:9, 9:16, 1:1, 4:3, 3:4, 9:21) and prompts up to 10,000 characters. It supports a fixed-camera toggle for locked-off shots. Best for simple, straightforward scenes where speed and cost matter more than fine motion detail. Add audio in post-production.

ByteDance Seedance Pro (V1 Pro) is a higher-quality text-to-video and image-to-video model at 480p, 720p, and 1080p in 5s or 10s clips, with seven aspect ratios (21:9, 16:9, 4:3, 1:1, 3:4, 9:16) and prompts up to 10,000 characters. It renders sharper detail and more consistent motion than Seedance Lite for the same duration range, making it a solid balance of quality and cost for marketing and product content. Add audio in post-production.

ByteDance Seedance Pro Fast is the accelerated image-to-video variant of Seedance Pro — same 480p/720p/1080p resolution options and 5s/10s durations, tuned for faster turnaround. It's a good fit for agencies and high-volume workflows that need Pro-level output without the wait. Add audio in post-production.

Kling 2.6 is Kuaishou's short-form text-to-video and image-to-video model with native audio generation — toggle the "sound" parameter to generate synchronized speech, effects, and ambience in a single pass. It generates 5s or 10s clips across 16:9, 9:16, and 1:1 aspect ratios, with prompts up to 1,000 characters. It's also the base model for Kling Motion Control, which transfers motion from a reference video onto a static image at 720p or 1080p. Best for social-media clips and dialogue-driven scenes with strong action physics.

Kling 3.0 is Kuaishou's latest video model with three quality tiers — Standard (720p), Pro (1080p), and native 4K — plus native audio generation, stronger action physics, and improved temporal consistency. It supports single-shot clips from 3–15 seconds and a multi-shot mode (up to 5 shots, each 1–12s) with per-shot prompts and @element references for consistent characters, objects, or audio across shots. Use natural-language prompts describing subject, motion, scene, and camera.

Kling 2.1 Master is the enhanced Kling 2.1 variant with stronger prompt adherence and higher visual consistency. It outputs 1080p only and is best for final-quality outputs. It excels at realistic motion, scene consistency, and cinematic camera control. Use natural language and cinematic camera terms. Negative prompts are supported. You can generate audio and merge it with the video in post-production.

Kling 2.1 Pro is a high-quality text-to-video and image-to-video model known for realistic motion, scene consistency, and strong camera control. It excels at action sequences, nature and landscape videos, and scenes with natural physics and movement. Structure prompts as: Subject → Movement → Scene → Camera / Lighting / Atmosphere. Use natural language and cinematic camera terms (close-up, tracking shot, dolly). Supports prompts up to 2,500 characters. Negative prompts are supported. Audio must be added in post-production.

Kling 2.1 is a high-quality text-to-video and image-to-video model focused on realistic motion, scene consistency, and strong camera control. It produces visually coherent videos with natural physics and movement, and works well for action sequences, nature and landscape content, and scenes that need smooth motion. It supports cinematic camera language (pan, dolly, tracking, handheld) and negative prompts, and works best with clear, natural-language prompts that describe subject, movement, scene, and lighting. Kling 2.1 does not generate audio; add music or voiceover in post-production.

Kling 2.5 Turbo Pro is Kuaishou's premium turbo model, generating 1080p-only clips at 5s or 10s across 16:9, 9:16, and 1:1 aspect ratios. It exposes a CFG (prompt-adherence) scale from 0–1.0 for fine-tuning how closely the output follows your prompt, and accepts prompts and negative prompts up to 2,500 characters. Best for premium advertising and brand content where consistent, controllable motion matters. Audio is added in post-production.

MiniMax H3 generates video from a text prompt, from a start and/or end frame, or from a mix of reference images, videos, and audio. It runs 4 to 15 seconds at 768P or 2K across six aspect ratios (21:9, 16:9, 4:3, 1:1, 3:4, 9:16), takes prompts up to 7,000 characters, and accepts up to 9 reference images, 3 reference videos, and 3 reference audio clips in reference mode. It suits longer clips that need to stay on-model with existing footage or artwork.

Runway is a high-fidelity AI video model for cinematic motion, smooth transitions, and precise camera control. It supports text-to-video and image-to-video, plus a Video Extend mode to lengthen an existing clip. 720p supports both 5s and 10s clips; 1080p is limited to 5s. Best for product reveals, atmospheric scenes, FPV fly-throughs, and tracking shots. Use direct, descriptive prompts that define motion and camera behavior. Does not generate audio.

Seedance 1.5 Pro is a motion-focused text-to-video and image-to-video model built for dynamic scenes, camera movement, and action. It supports multi-shot sequences with "Shot Switch" and optional AI-generated audio (sound effects and ambience) in one pass. Use motion-first prompts: subject, action, and camera. Adverbs like "slowly" or "quickly" control motion intensity. Negative prompts are not supported.

Veo 3.1 Fast is Google's cinematic AI video model built for high-fidelity motion, precise camera control, and structured storytelling, with 720p, 1080p, and native 4K output. It suits cinematic storytelling, product demos, educational content, and scene-based narratives. Describe the scene and environment, specify actions, and add camera movement (pan, dolly, static). Think in scenes, not single frames. More detail generally improves results.

Veo 3.1 Lite is Google's efficient Veo variant — the model included on the Free plan. It generates cinematic motion with native audio at 720p and 1080p in 4, 6, or 8-second clips, making it ideal for quick drafts, social clips, and trying ideas before committing credits to Fast or Quality.

Veo 3.1 Quality is Google's premium cinematic AI video model for high-fidelity motion, precise camera control, and structured visual storytelling, with 720p, 1080p, and native 4K output. It suits cinematic storytelling, product demos, educational content, and scene-based narratives. Describe the scene and environment, specify actions, and add camera movement (pan, dolly, static). Use temporal language (slow, deliberate movement) for more natural motion. Clarity and intent matter more than length.

WAN 2.2 is a text-to-video and image-to-video model with improved motion fidelity, prompt adherence, and cinematic lighting. It is tuned for short, emotion-driven videos—joy, calm, healing, inspiration—where mood, motion, and atmosphere matter more than long-form narrative. Use clear visual details and atmosphere words (warm, gentle, cinematic) for best results.

WAN 2.5 is a fast text-to-video and image-to-video model aimed at beginners and high-frequency creators. It produces short-form videos suited for social media, marketing, and educational content. Keep prompts simple and descriptive. Best for TikTok, Instagram, and YouTube Shorts. Add narration or music in post-production.

Wan 2.6 is Alibaba's structured-storytelling video model, generating text-to-video and image-to-video clips at 720p or 1080p in 5, 10, or 15-second durations from prompts up to 5,000 characters (Chinese and English supported). It's optimised for deliberate pacing and strong prompt adherence — well-suited to product demos, brand videos, and educational explainers that need longer, more structured scenes than Wan 2.5.

Wan 2.6 Video-to-Video applies a new text prompt to an existing reference video (up to 3 clips, 10MB each) while preserving the original structure and timing — useful for style transfer and re-skinning footage you already have. Output is 720p or 1080p at 5s or 10s, with prompts from 2–5,000 characters in Chinese or English.

Wan 2.7 is Alibaba's latest text-to-video and image-to-video model, supporting 720p and 1080p output from 2 up to 15 seconds — the widest duration range in the Wan lineup — across five aspect ratios (16:9, 9:16, 1:1, 4:3, 3:4). It accepts prompts up to 5,000 characters plus an optional negative prompt, and can take a custom audio track alongside the generated video. It suits longer-form, structured storytelling and benefits from clear visual detail and atmosphere words.