Generate video with Veo, Sora, Kling, Runway, Hailuo, Wan, Hunyuan, Grok Imagine, and every other major model. Pay per second, no subscription.
18 models supported · pay-per-credit · credits never expire
Hand-tested star ratings against credit cost, one dot per model family at the tier we recommend. The leftmost dot in each row is the cheapest way to that quality tier. Hover a dot for its name, click it to jump to that model below.
Every model here runs on the same pay-per-credit balance, so you can switch freely. If you just want a starting point:
The default model in the Upsampler app: the best balance of quality, speed, and price. A post-trained MiniMax H3 with tighter prompt adherence, at half the price.
See detailsBest current placement on the Artificial Analysis arena. MiniMax's frontier video model, among the strongest on the Artificial Analysis arena.
See detailsSolid results at one of the lowest credit prices here. 4x faster draft previews for rapid iteration before committing to a full render.
See detailsPrompt: Astronaut sprinting on the moon
The newest revision of xAI's stylized video generator — an image-to-video model with improved motion and lip-sync alongside synchronized audio. Keeps the distinctive xAI personality, tuned for X-platform-aligned and stylized social content where tone matters more than polish.
Prompt: The camera follows the man in black as he flees quickly, followed by a group of people chasing him. The camera switches to side tracking. The character panics and knocks down a fruit stand on the roadside, gets up and continues to run away. The crowd makes panicked sounds.
A post-trained variant of MiniMax H3, tuned for stronger prompt adherence and cleaner aesthetics while co-optimized with fal's custom inference stack for higher throughput, at half the base model's price. It keeps everything H3 is good at: unusually stable subjects through long takes, clips out to fifteen seconds where most flagships stop at eight, and first-and-last-frame control for transitions. Served at 768P, with no audio track, so plan on scoring these clips in the edit.
Prompt: Astronaut sprinting on the moon
xAI's stylized video generator with synchronized audio. High-quality text-to-video and image-to-video with the distinctive xAI personality. Strong for X-platform-aligned content and stylized social video where polish matters less than tone.
Prompt: A woman dancing in a garden full of animals. She is wearing a T-shirt with the word “Upsampler” on it.
Example from Veo 3.1
Google DeepMind's Veo 3.1 Fast with synchronized audio. Industry-leading dialogue lip-sync and audio realism — context-aware audio generation, smooth motion, and native video and audio extension. Reach for it on dialogue-heavy short-form content where Veo's lip-sync advantage justifies the cost.
Prompt: A woman dancing in a garden full of animals. She is wearing a T-shirt with the word “Upsampler” on it.
Example from Seedance 1.5 Pro
ByteDance Seedance 1.5 Pro with synchronized audio. Cinema-quality video with precise lip-syncing and cinematic camera control — strong for narrative short-form content, music video aesthetics, and anywhere dialogue matters as much as visuals.
Prompt: A woman dancing in a garden full of animals. She is wearing a T-shirt with the word “Upsampler” on it.
Lightricks' fastest LTX 2.3 tier with synchronized audio. Open-weight cinematic concept iteration — describe a clip with audio cues, get a quick render with sound. Use for iteration before stepping up to LTX 2.3 Pro for final renders.
Prompt: A corgi in sunglasses cruising on a skateboard past a taco stand in slow motion, supremely confident, hip-hop beat with a record scratch as it nods at the camera
Pruna AI's production-tier video model with fast generation, built-in audio, and multi-aspect-ratio support. Optimized for cost-quality balance — solid for production workflows where the top-tier closed models (Veo, Sora) feel too expensive for the use case.
Prompt: Girl petting the dragon
Example from Wan 2.2 14B
Alibaba's flagship Wan 2.2 video model with crisp 480p output and strong stylization. Open-weight availability makes it useful for self-hosted pipelines and teams that want production-quality video generation without closed-API costs. Edged out by Veo, Sora, and Kling at the top tier but cost-competitive.
Prompt: The camera races along just above the asphalt beside a matte-black motorcycle weaving through slow traffic on a desert highway at dusk. The rider glances over his shoulder, drops a gear, and threads the needle between two trucks with inches to spare while dust kicks up into the golden backlight. Heat shimmer bends the horizon and the low sun flares across the lens.
MiniMax's frontier video model and one of the highest-Elo entries on both the text-to-video and image-to-video arenas at Artificial Analysis. It is unusually good at holding a subject stable through long takes, which is why it runs out to fifteen seconds where most flagships stop at eight. Served here at 768P to keep the per-second price in range, with first-and-last-frame control for transitions. It generates no audio, so plan on scoring these clips in the edit.
Prompt: The camera follows the man in black as he flees quickly, followed by a group of people chasing him. The camera switches to side tracking. The character panics and knocks down a fruit stand on the roadside, gets up and continues to run away. The crowd makes panicked sounds.
ByteDance Seedance 2.0 is the next-generation Seedance flagship with native audio, multimodal inputs, and 720p output. Among the leaders on the Artificial Analysis text-to-video arena. Reach for it on hero clips, premium ad creative, and narrative content where the cost per second is justified.
Prompt: A muscular boxer trains alone in a dim warehouse gym, driving rapid combinations into a heavy bag. Sweat flies off with every impact and the chains rattle as dust drifts through a single shaft of window light. The camera circles him slowly, then punches in hard on his final hook.
ByteDance's latest Seedance generation, running here at 480p so the flagship model stays affordable per second. It keeps the things Seedance is known for, native synchronized audio, strong physical motion, and dialogue that lands on the beat, and adds first-and-last-frame control for scripted transitions. Reach for it when you want Seedance 2 direction quality on a clip where the extra sharpness of a 720p render would not change the edit.
Prompt: Rain streaking across a train window as green countryside slides past, focus racks slowly from the droplets on the glass to a distant farmhouse, melancholic golden-hour light, cinematic and quiet
Runway Gen-4.5 is premium text-to-video and image-to-video with cinematic quality, rich detail, and fluid motion. Mature creator-focused tooling with strong cinematic handling. The natural choice for teams already on Runway's broader video stack.
Prompt: The camera follows the man in black as he flees quickly, followed by a group of people chasing him. The camera switches to side tracking. The character panics and knocks down a fruit stand on the roadside, gets up and continues to run away. The crowd makes panicked sounds.
Kling 2.5 Turbo Pro — the flagship Kling tier with pro-grade text-to-video and image-to-video. Smooth motion, strong prompt fidelity, and exceptional motion physics for complex camera work — tracking shots, dolly moves, and crane sweeps all hold up. Strong choice for cinematic narrative work.
Prompt: A woman dancing in a garden full of animals. She is wearing a T-shirt with the word “Upsampler” on it.
ByteDance's Seedance 1.0 — fast, cost-efficient video generation with strong motion physics and detail. The original Seedance flagship, now succeeded by Seedance 1.5 Pro and 2 Fast for top-tier work. Still a practical pick when 1.5 Pro and 2 are overkill for the brief.
Prompt: The camera follows the man in black as he flees quickly, followed by a group of people chasing him. The camera switches to side tracking. The character panics and knocks down a fruit stand on the roadside, gets up and continues to run away. The crowd makes panicked sounds.
Lightricks' flagship LTX 2.3 with synchronized audio. Higher-fidelity video generation in the open-weight LTX line, with cinematic camera handling and audio sync. Strong choice for teams that want premium video quality on infrastructure they control.
Prompt: Small chunky hippo sprinting towards the camera
The speed tier of Lightricks' LTX 2.5, generating 720p video with synchronized audio from either a prompt or a still. At three credits per second it is one of the cheapest audio-capable video models on the platform, and it stretches to 20 seconds in a single generation where most models stop at 8 to 12. That combination makes it the natural workhorse for long-form b-roll, social loops, and any storyboard pass where you want to see many variations before spending on a flagship. Note that it only outputs landscape or portrait video: a reference image in any other shape is center-cropped to whichever of the two fits it best.
Prompt: A woman dancing in a garden full of animals. She is wearing a T-shirt with the word “Upsampler” on it.
Lightricks' distilled LTX 2 — open-source audio-video model for expressive clips with sound. Lower fidelity than the full LTX 2 / 2.3 line but practical for open-weight workflows and teams that want LTX-style video without closed-API costs.
Prompt: A woman dancing in a garden full of animals. She is wearing a T-shirt with the word “Upsampler” on it.
Pruna AI's draft-tier video generation — roughly 4x faster than the full P-Video model for rapid previews. Built for the iteration phase: nail down the motion and composition cheaply, then commit to a full render only when you're sure. Cuts cost meaningfully on exploratory work.
Arena scores and rankings by Artificial Analysis and LMArena (leaderboard dataset, CC BY 4.0).