| Last Updated | Changes |
| 11/14/2024 | First Version |
| 12/17/2024 | New Video Generation options |
| 1/15/2025 | Bring Your Own Image – Img2Video From Any Image! |
| 2/5/2025 | Lightricks LTXV On-Site! |
| 4/17/2025 | Vidu, Hunyuan and Wan 2.1 On-Site! |
Text2Video & Image2Video in the Generator?
We’re thrilled to announce a major new feature in the Civitai Generator – video generation! You can now create short video clips directly from text prompts, or by using a “starter” image. This exciting addition opens up fresh creative possibilities, making it easier than ever to bring your ideas to life in motion!
Over time, we’ll be adding more video generation tools to give you even more style options, control, and flexibility in your video creations. As we continue to expand, you’ll notice new tools, interface tweaks, and enhancements that make creating and customizing your videos more intuitive and powerful. Stay tuned for updates as we refine and grow this feature based on your feedback and our ongoing improvements and partnerships!
Note: For detailed information on navigating the Civitai Generator’s image creation interface, see our Guide to the Civitai Image Generator.
Tools
We currently have eight tool choices for video generation, with more in the pipeline!;
Haiper
Haiper.ai’s 2.5 model, is a robust and advanced engine designed for producing both text-to-video and image-to-video content. This model brings a seamless, straightforward approach to video creation, requiring minimal setup to achieve impressive results.
Haiper.ai offer a range of helpful resources to ensure you get the most out of their video generation service. Since prompting for text-to-video can be a bit different than what we’re used to with text-to-image models on Civitai, these resources are designed to guide you through the nuances of video-focused prompts.
Hailuo by MiniMax
Hailuo, developed by the Chinese company MiniMax, is an innovative AI video generator that transforms text prompts and images into dynamic, high-quality videos. Launched in September 2024, it allows users to create videos using both text-to-video and img-to-video workflows, with a range of aspect ratios supported.
Kling
Kling AI, developed by Kuaishou Technology, is a cutting-edge text-to-video and image-to-video generation model that enables users to create high-quality videos from text descriptions or static images. Launched in June 2024, Kling can produce videos, accurately simulating real-world physics and complex motions. Its features include advanced camera controls, motion brushes for precise object movement, which aren’t available on Civitai.com’s Generator just yet, but we’re working on those!
Lightricks LTXV
Lightricks LTX-Video is the first DiT-based video generation model capable of generating high-quality videos in real-time. It produces 24 FPS videos at a 768×512 resolution practically faster than they can be watched! It works in text-to-video and image-to-video modes, and it’s the first video service running entirely in-house, on our own GPU hardware!
Mochi
Mochi 1 preview, by creators Genmo, is an open source, state-of-the-art, video generation model with high-fidelity motion and strong prompt adherence. Along with on-site Generation, Mochi can be used offline/locally, with ComfyUI and at least 12 GB of VRAM!
You can read our full Quickstart Guide to Mochi, here!
Vidu
Vidu delivers cutting-edge text-to-video generation, enabling users to create dynamic, visually compelling videos from simple prompts. Whether you’re aiming for cinematic flair, artistic stylization, or unique visual effects, Vidu offers a wide range of high-quality output styles to suit your creative needs. Support for image-to-video generation is also on the horizon! Check out Vidu’s prompt guide for more information on how to get the best out of this phenomenal model!
Hunyuan
Hunyuan is Tencent’s powerful text-to-video generation model, now available on our platform with full video LoRA support. This allows for greater flexibility and fine-tuning of visual outputs to better match specific styles or themes. While currently offered in text-to-video mode, we are working to enhance fidelity, improve overall quality, and accelerate generation speeds to deliver an even better experience.
Wan 2.1
Wan 2.1 is Alibaba’s foray into text-to-video technology, offering visually rich and coherent video outputs from natural language prompts. This model emphasizes strong scene consistency, smooth motion, and detailed artistic styles. It’s an excellent choice for creators looking to produce high-quality videos with minimal input. We’re continuing to monitor performance and explore expanded functionality – it’s not a lightweight model and requires significant hardware resources! We’re currently using the 480p version rather than the 720p, but this will be revised as we adjust our pipeline.
Creation Options
While each tool may have a unique interface with various options and settings, the core functions of Video Generation are as follows:
Text2Video
There is now a Video modality tab at the top of the Generator. This gives access to all of the Video Generation functions.
- Tool – The Tool drop-down switches between the available video generation “engines”.
- Workflow – Current workflows include Text to Video (the default) and Image to Video (see below).
- Prompt (Positive and Negative) – self explanatory; specify what kind of motion you want, and don’t want to see in your output.
- Aspect Ratio – Some video generation Tools offer a range of image dimensions.
- Duration – Some video generation Tools offer a range of video durations.
- Seed – Much like image generation, video generation may use a seed value, to reproduce specific results.
- Prompt Enhancer – Some Tools may provide a prompt-enhancement service to beautify and enhance your prompts prior to submission. You won’t see the result of this enhancement.
- Additional Resources – Some Tools (Hunyuan, Wan) offer the ability to apply video LoRAs to generations! These can be added from the “Add” button, much like a text2img generation.
Image2Video
Some tools let you use a “starter” or initialization image as the starting point for a video.
To try this, pick any image from your Generation Queue or Feed, then choose the “Image to Video” option from the context menu, or the Magic Wand menu, shown on any image.
The image will be loaded into the Image to Video interface, where a prompt can be provided, specifying the desired movement.
There are multiple video tools with a capacity for image to video – when an image is loaded into the interface, you can select which you’d prefer to use from the dropdown list;
Bring Your Own Image – Img2Video From Any Image!
Bring Your Own Image (BYOI) lets you easily import images for our img2video services.
Drag images directly from Civitai, other websites, or your local PC into the drop-zone, or use an image URL. Supported formats include .png, .jpeg, .jpg, and .webp, with a maximum file size of 16 MB.
Images uploaded to the generator via BYOI are not posted to your Civitai Profile.
Be aware that uploaded images may be processed off-site by third-party video generation API providers and are subject to their policies and Terms of Service. Each service has its own content rules. If your request is denied by the partner API, you’ll receive a notification, and your Buzz will be refunded.
Output Examples from the Civitai Generator
Limitations
- This is the first iteration of video generation to feature on Civitai.com, and as such it may lack some of the polish of other generation features! We’ll be improving video generation over time by tweaking the interface, introducing new quality and speed options, and implementing new Tools.
- Pricing! We know video gen is pricey; it’s cutting-edge, resource-demanding technology. We’re actively working with our inference partners to explore options that make it more accessible for everyone.
- Terms of Service. While generating videos, you must adhere to Civitai’s standard Terms of Service. Additionally, please ensure you follow the Terms of Service specific to the tool you’re using for video generation.
- Results. Video generation can be unpredictable – getting the perfect output may take a few tries, as each result can vary, widely. Please note that we cannot refund generations that complete successfully but don’t meet your expectations.
Additionally, once a generation begins processing, it’s locked-in and cannot be canceled. You might be able to cancel immediately after starting, but once generation is underway, it’s committed. Of course, if the generation fails, refunds will be provided.
- Provider instability. Since many of the video options available via the Civitai Generator depend on external API partners to handle video generation requests, occasional failures, outages, or disruptions may occur that are outside of Civitai’s direct control.
- Kling Generation may return a “No Provider Supports this Job” error which is related to their API and not something we have control over. Jobs which return this error will be refunded instantly.
- Hailuo by Minimax does not have the concept of a seed. If running the same prompt, or the same input image with the same prompt, you’ll receive the same output. You will not be charged for this duplicate output.
- Hailuo by Minimax and Kling have relatively small concurrency limits; the number of jobs we can send to their API at the same time. At peak times, there may be a queue!
I need more help!
If you’re experiencing issues generating Video with the Civitai Image Generator and a solution isn’t mentioned on this page, please reach out to our Support Team at [email protected].
