A node-based creative canvas, video models we optimize ourselves, and one API for the major cloud models from Alibaba Cloud, Tencent Cloud and others. Use it pay-as-you-go, or run the whole stack on your own GPUs: lower prices, faster renders, and your data stays yours.
One platform — canvas, models, API, scheduling and quotas — configured and deployed for how you work.
A ready-to-use AI video environment for media, film and animation programs, and a platform research groups can run model experiments on.
Product films, e-commerce clips, training and explainer videos: build the flow on the canvas once, then produce in batches.
Go from script to storyboard, previs and concept footage in one place. Faster pitches, and short dramas or ads can ship straight from the canvas.
A node-based canvas lays the whole production out in front of you: which prompt, which model and which reference made each shot — traceable, and re-doable shot by shot.
Wire script, shot list, image, video and audio nodes freely; save flows as templates.
A shot list splits into shots automatically; render stills and video in batches with live progress.
Ask in plain language and the assistant builds flows, edits shots and swaps models on the canvas.
A vision model checks every shot, speech recognition checks the lines, and failed shots re-render automatically.
Built-in speech synthesis with custom voices per character, in sync with picture.
Timeline editing, subtitles and music; export the final cut in one click.
We tune the self-hosted models ourselves and proxy the cloud vendors' models. Canvas nodes and the API use the same call format, so switching models is one parameter.
Run on dedicated GPUs; tunable, accelerated, deployable on-prem.
No separate vendor accounts, integrations or top-ups: one account, one API, one invoice.
Cloud models are billed per use at 20% off the official list price; high monthly volume or annual contracts get tiered discounts. Dedicated deployments are priced monthly by GPU, or as a one-time delivery.
| Model | Spec | Unit | Official list | Our price | Save |
|---|---|---|---|---|---|
| Our optimized models · self-hosted | |||||
| MiniMax H3 · optimizedOURSvs. MiniMax official API | 768P · text/image/ref-to-video · with audio | ¥ / sec | 0.50 | 0.40 | −20% |
| Cloud video models · unified API | |||||
| Wan 3.0Alibaba Cloud | 480P · text/image-to-video | ¥ / sec | 0.30 | 0.24 | −20% |
| Wan 3.0Alibaba Cloud | 720P · text/image-to-video | ¥ / sec | 0.60 | 0.48 | −20% |
| Wan 3.0Alibaba Cloud | 1080P · text/image-to-video | ¥ / sec | 1.20 | 0.96 | −20% |
| Hunyuan Video 1.5Tencent Cloud | 720P | ¥ / sec | 0.30 | 0.24 | −20% |
| Seedance 2.5Volcano Engine | 720p · from official 5s example | ¥ / sec | 1.51 | 1.21 | −20% |
| Seedance 2.5Volcano Engine | 1080p · from official 5s example | ¥ / sec | 3.74 | 2.99 | −20% |
| Seedance 2.0 FastVolcano Engine | 720p · from official 5s example | ¥ / sec | 0.80 | 0.64 | −20% |
| Kling 3.0Kling | 720P · no audio | ¥ / sec | 0.60 | 0.48 | −20% |
| Kling 3.0Kling | 1080P · no audio | ¥ / sec | 0.80 | 0.64 | −20% |
| Hailuo 2.3MiniMax | 768P · per 6s clip ÷ 6 | ¥ / sec | 0.333 | 0.266 | −20% |
| Hailuo 2.3MiniMax | 1080P · per 6s clip ÷ 6 | ¥ / sec | 0.583 | 0.466 | −20% |
| Cloud image models · unified API | |||||
| Qwen-Image 3.0Alibaba Cloud | 1K / 2K | ¥ / image | 0.18 | 0.144 | −20% |
| Wan 2.7 ImageAlibaba Cloud | text-to-image | ¥ / image | 0.20 | 0.16 | −20% |
| Hunyuan Image 3.5Tencent Cloud | 1K / 2K (preview) | ¥ / image | 0.15 | 0.12 | −20% |
Official list prices are taken from each vendor's public pricing page (checked 2026-11-11); the vendor's current price applies, promotions excluded. Our prices are standard pay-as-you-go rates in RMB; high volume is quoted separately. Pricing for our self-hosted models depends on the deployment.
Off-the-shelf models rarely match your characters, look and camera language. We tune them on your material, and keep cutting render time.
LoRAs for characters, art styles, brand elements, specific motions and camera language. Training data and the resulting weights are used only in your deployment.
Every tuning round renders A/B comparisons on a fixed test set. You sign off on the improvement before it ships.
Same quality, shorter waits: attention optimization, quantization, multi-GPU parallelism and two-pass HD, combined to fit your hardware.
Attention acceleration alone cut render time by about 18% in our tests (same GPU, prompt and resolution). Gains from quantization and multi-GPU depend on hardware; we benchmark on your machines before deployment.
One platform, three ways to deploy. Data-sensitive teams can go fully on-prem and keep both compute and data in-house.
Call our API and use the web canvas directly. Nothing to deploy; good for trials and smaller workloads.
A GPU pool set aside for you, with its own platform instance and database, in the region you choose. We run it.
Delivered as container images to your servers and run inside your network. Weights are distributed encrypted; upgrades remotely or on site.
Your use cases, volume, data requirements and current hardware.
A trial account, and a batch of samples from your real briefs.
Custom LoRAs, then deployment and load testing in the mode you chose.
Monitoring, model upgrades and speed work, with a named contact.
No. Your uploads, prompts and outputs are used only for your own jobs. LoRA training runs only on data you authorize, and the weights are used only in your deployment.
Requests are forwarded to that cloud vendor and fall under its terms. For highly sensitive work, use the self-hosted models or an on-prem deployment.
A standard REST API with async jobs: submit a job, then poll or receive a callback. Self-hosted and cloud models share one parameter format; switching is just the model name.
Pay-as-you-go has no minimum; top up and go. Dedicated cloud is billed monthly; on-prem is quoted per project.
The platform moderates prompts and frames, and has a reporting and takedown process. Illegal content is not supported. On-prem deployments can plug in your own moderation policy.
48 GB+ VRAM per GPU is recommended for video models, and multiple GPUs run in parallel for speed; smaller cards can use quantized builds. Tell us what you have and we'll benchmark and propose a setup.
Leave your contact details and a rough brief, and we'll get back to you shortly to set up a demo and trial.
Or email us directly:
contact@cinecompute.com