Runway Gen-3 video generation tutorial, 2026 AI short film from script to finished film

Runway Gen-3 is the third-generation AI video generation model launched in July 2024. It will be upgraded to the Gen-3 Alpha Turbo version in November 2025, and Gen-4 will be launched in April 2026. However, Gen-3 is still the most cost-effective workhorse. It only takes 70 seconds to generate a 10-second video. The two indicators of 1080p image quality and motion coherence surpass Sora 1.0 and are above Pika 2.0 and Keling 1.6.

This article summarizes the complete process of Runway Gen-3 from registration to production. It includes six complete processes including account opening, Vincent video, Tusheng video, video to video conversion, sound synchronization, and editing and exporting. Finally, let’s give a practical example of a complete short film and how long it took from the script to the 60-second film.

Runway is a product made by what team?

Runway was founded in 2018 by Cristobal Valenzuela and is headquartered in New York. It is currently valued at $3 billion. Investors include Google, Nvidia, and Salesforce. The team's earliest early participants in the image generation model Stable Diffusion came from Runway's internal research group.

The Runway product line includes video generation Gen series, image generation Runway ML, and sound generation Sonic. Video is the core, and Gen-3 has been used by media companies such as Netflix, Disney+, CNN, and A24 for editing special effects and short film creation. 30% of Netflix documentary trailers in 2025 will be generated by Runway.

Differences between Gen-3 and Sora. Sora's long video limit is 1 minute, which is more physically realistic but slower to generate. Gen-3 has a maximum limit of 10 seconds but is fast to generate, suitable for rapid iteration of short films. Gen-3 has the highest usage rate in the short video e-commerce Douyin content creation circle.

Account registration and subscription plan

Open runwayml.com and click Sign Up. Registration supports Google, Apple, and email. The international version of the mobile phone number can be used to register overseas credit cards or PayPal recharge. Mainland China accounts can be registered, but subscriptions require overseas payment methods.

There are 5 subscription plans. The free version’s 125 credits per month is enough for about five 10-second videos. Standard monthly fee is $12, 625 credits per month. Pro costs $28 per month for 2250 credits per month plus 4K export. Unlimited costs $76 per month to generate unlimited amounts but requires queuing. Enterprise custom price is exclusive to large customers.

The best deal for newbies is Standard $12 a month. You can make 25 10-second videos, which is enough for testing and daily social media creation. If it is a commercial project that requires 4K, go directly to Pro. Unlimited does not recommend paying back the money unless you produce more than 5 videos per day.

50% off student package requires .edu email verification. Developers provide 100 credits to try out the API for free.

Vincent video core operation

After logging in, click Generate Video to enter the workbench. On the left is the Text to Video input box, in the middle is the preview, and on the right is the parameter panel.

There are three key elements to writing prompt. The subject description is clear, such as "an orange tabby cat stretching in the sun", the shot description is close up, wide shot, tracking shot, and the style keywords are cinematic, anime style, and photorealistic.

The first prompt is actually tested. Enter a cyberpunk city at night neon lights reflecting on wet streets cinematic 8 credits and output after 80 seconds. The picture shows the neon reflection of the cyberpunk city on the wet street at night, and the camera slowly pans up from the ground, which matches the description of the prompt.

The parameter panel has 3 key settings. Duration is 5 seconds or 10 seconds, 5 seconds consumes 5 credits, and 10 seconds consumes 10 credits. Aspect Ratio 16:9 for horizontal screen Douyin 9:16 for vertical screen 1:1 for square screen. After the Seed seed value is locked, the result is similar to prompt multiple generation.

Tusheng video makes the picture move

In addition to Vincent videos, Tusheng videos are more practical. Start with a static image and make it move.

The uploaded image can be a photo taken with a mobile phone and generated by Midjourney DALL-E, and any proportion is acceptable. Runway will automatically adapt. Click Image to Video, drag in the image, and enter prompt to describe the part you want to move.

Actual test case. A picture of a Ghibli-style girl standing in a field of rapeseed flowers, prompt reads wind blowing through her hair flowers swaying gently camera slowly orbiting around her. The generation is completed in 10 seconds. The hair is fluttering and the petals are gently panned and the camera slowly rotates 270 degrees around the protagonist, which is comparable to real shooting.

Tusheng videos are 5 times more controllable than Vincent videos. First use Midjourney to create a satisfactory static picture, and then use Runway to make it move. It is suitable for scenes that require precise control of picture details. This is the standard workflow for professional creators.

Motion Brush local motion control

The Gen-3’s killer feature is Motion Brush. After uploading the image, use a brush to smear the designated area on the screen. Only the smeared part will move and other parts will remain stationary.

The first application scenario is product advertising. For example, in a product photo of a pair of sports shoes, only the soles are painted with Motion Brush and then the prompt "shoe sole bouncing on ground" is entered. The generated video only has the Q-bouncing effect on the soles and the other parts are stable, and the texture is cleaner than the real shot.

The second application scenario is the animation of emoticons. Use Motion Brush to paint a cat emoticon on the tail and type "tail wagging slowly" to generate a slow wagging tail effect and add text to it and send it directly to the Moments group.

Motion Brush provides 10x more precise control than plain text. Only after learning this function can Runway really open up. The free version can only use Motion Brush 5 times per month, and the Standard plan has unlimited use.

Sound and music synchronization

After the video is generated, it will be silent by default and music and sound effects need to be added. Runway's built-in Sonic Soundtrack library has more than 500 copyright-free soundtracks, classified by mood as suspenseful, upbeat, epic, and soothing. Click Add Audio, select a song and drag it to the timeline length to automatically match the video.

Even more advanced is AI sound effect generation. Click Generate Sound Effects and enter "footsteps on gravel" or "thunder rumbling" to generate a corresponding sound effect in a few seconds. Can be added to any video segment.

Use the Lip Sync function for narration. Upload a narration audio you recorded, and Runway will automatically recognize the character's mouth shape in the video so it can lip-sync with your voice. Video processing time in 10 seconds is 30 seconds, and the results are very natural in medium-brightness lighting.

To export the finished film, choose MP4 or MOV format. 1080p standard 4K requires Pro package. After downloading, you can directly tweet to YouTube and Instagram. Runway does not add watermarks.

API access for batch automation

You need to programmatically generate large batches of videos using the Runway API. The monthly Pro package comes with API quota. You can visit developer.runwayml.com to obtain the key.

The API interface uses Python SDK. pip install runway and then import from runwayml import Runway, initialize client. Call client.image_to_video passing in the image_url and prompt_text parameters to return task_id. Poll client.tasks.retrieve(task_id) until the status is SUCCEEDED to get the output_url and download it.

In batch scenarios, for example, an e-commerce company requires 100 product animations for 100 items. The script calls image_to_video in a loop to run serially for a single 80 seconds. The Pro plan costs 2250 credits per month and can run 225 10-second videos.

The API current limit allows 3 concurrent tasks to be queued for a single account. It is recommended that batch tasks be serialized with a sleep interval of 5 seconds to avoid task backlog.

Complete short film case 60 seconds science fiction short film

Actual measurement of a complete process. The goal was to make a 60-second cyberpunk detective short.

The first step is the storyboard. Break 60 seconds into six 10-second shots. Shot 1 Panoramic view of the city at night. Shot 2: Close-up of the protagonist pushing open the bar door. Shot 3 The bartender pours the wine. Shot 4: The protagonist answers the phone with a solemn expression. Shot 5 The protagonist walks out of the bar street. Shot 6 freezes the back view in the distance.

The second step is to generate static storyboards. Use Midjourney to generate a satisfactory picture for each shot, and add "cyberpunk noir detective movie still cinematic" to the unified style of the keyword. 15 minutes for 6 pictures.

The third step is to use Runway to generate videos. Add a prompt to each picture to describe the movement method for 10 seconds. 6 shots total 60 credits 8 minutes.

The fourth step is combination. Download 6 videos and import them into CapCut or Premiere to splice and add transitions.

The fifth step is to add music. Select a cyberpunk noir soundtrack from the Runway Soundtrack library and drag it to the timeline.

The sixth step is to add dubbing. Generate the protagonist's inner monologue narration audio import track using ElevenLabs or Suno.

The total time is 60 minutes and the cost of a 60-second film is US$12, which is affordable with the Standard monthly fee. The traditional shooting cost of the same finished film is at least 50,000 yuan.

What kind of content creator is it suitable for?

The bloggers who make short videos on Douyin Xiaohongshu Station B will benefit most directly. The 10-second video is generated by high-quality AI, and combined with the narration, you can choose popular topics such as "AI-generated cyberpunk world" and "I let AI act out my dreams".

E-commerce sellers directly use Tusheng videos to create product animations. A product photo turned into a dynamic ad has a 50% higher CTR than a static image. Taobao’s TikTok store owner picture video requires 5 to 9 seconds and Runway is perfectly adapted.

Advertising creative agency uses Runway for proposals. When clients have meetings, they don’t have to wait for editors to make demos. Designers can use Runway to come up with concept film demonstration directions on the spot, improving communication efficiency by 5 times.

Independent filmmaker. Short film production, low-cost trial shooting effects film to test the lens language. If you are not sure how to shoot a script, you can first use AI to generate a reference film and then go to the actual shooting.

Limitations and things that cannot be done temporarily

The first long video is capped at 10 seconds. If it exceeds 10 seconds, it needs to be spliced, but the consistency between different paragraphs is poor, and the character's appearance will change. This is a common bottleneck for all AI video models in 2026 and is expected to break through in 2027.

The second complex action. Multi-joint fast movements such as fighting, parkour, and dance often cause deformation. Slow motion and still shots work well.

Third text rendering. Text signs, subtitles, and logos that appear in videos are often garbled or distorted. Runway 4.0 has improved but is still unreliable. Commercial scene text needs to be added in post-PS.

Fourth physical violation. Physical effects such as water flow, fire, and glass shattering are occasionally unintuitive, such as water flowing backwards and glass pieces floating.

Fifth copyright risk. Runway training data has not been fully disclosed, and the generated videos have copyright disputes in Europe and the United States. It is recommended to read the "Indemnity" compensation clause in the product terms before commercial use.

FAQ

Can Runway be used normally in mainland China?

Yes but requires a stable international network. Access to the Runway server in the United States and Europe requires a bandwidth of more than 100Mbps, otherwise uploading images will time out frequently. The generated video is downloaded in 1080p 10 seconds and is about 50MB, which is barely enough for domestic mobile networks. Subscription requires an overseas credit card or PayPal, domestic credit card, Visa Mastercard logo, you can try, but there is a 95% chance that it will be risk controlled. It is recommended to bind overseas relatives and friends cards or buy virtual cards. The free version can be registered with a domestic email address to experience it, but it cannot be upgraded after 125 credits are used up.

How long does it actually take to generate a 10-second video?

A period of 60 to 90 seconds under normal load. Standard and Pro plans have priority queues with less waiting. The queue for the free version lasts 5 to 15 minutes between 8pm and 11pm during peak hours. Fastest average 50 second segment from 6am to 9am. Batch generation of 10 videos actually takes 15 to 25 minutes. The Unlimited package is always at the end and has the lowest priority, so it is generally not recommended.

What is the difference between Runway Gen-4 and Gen-3? Should I upgrade?

Gen-4 is scheduled to be released in April 2026 and focuses on physical realism and lens consistency. In the same prompt, the quality of the video is improved by 30%, but the credits consumed per time are doubled, and the number of videos that the Standard package can generate is halved. Gen-3 is more cost-effective for daily social media, and Gen-4 is used for commercial projects and high-quality requirements. The two models can be switched in the generation interface.

How to improve film production quality to commercial level

Four tips. The first thing to do is to use pictures to make videos, instead of just using text to make videos. First use Midjourney to make high-quality storyboards. The second prompt uses cinematic shot on Arri Alexa film grain and other movie texture keywords. The third Motion Brush finely controls the motion area to avoid overall picture shake. Fourth, the same prompt is retried 3 to 5 times to select the most satisfactory version. AI video generation is naturally random and has a 30% chance of being satisfied once.

Can you calculate the total cost per year with the price?

Estimated based on generating 10 10-second videos per month. Standard is US$12 per month, US$144 per year, which is approximately RMB 1,050. Each video costs $1.2 8.6. If the same quality is used, a traditional outsourced editing company will quote 200 yuan for a 10-second video. Runway saves 95% a year. This is an irreversible tool for small and medium-sized content teams and individual creators.