Runway Gen 4.5 Tutorial: How to Create AI Videos Step by Step
Runway Gen 4.5 currently sits at the top of the Artificial Analysis Text to Video benchmark with 1,247 Elo points. We spent three weeks generating over 200 clips across text-to-video and image-to-video modes, testing everything from simple product shots to complex multi-character scenes. This tutorial covers exactly how to use it, what it costs, where it genuinely impresses, and where you'll burn credits on mediocre output.
TL;DR — Key Takeaways
- 1Gen 4.5 delivers the most physically accurate AI video available right now—realistic weight, momentum, fluid dynamics, and fabric textures that previous models couldn't handle.
- 2Two generation modes: text-to-video (describe the scene from scratch) and image-to-video (animate a static image). Image-to-video is consistently more reliable.
- 3Credits burn fast: 25 credits per second. The Pro plan's 2,250 credits gets you about 90 seconds of Gen 4.5 footage—around nine 10-second clips per month.
- 4Known weaknesses: no native audio, occasional object disappearance mid-frame, and effects sometimes precede their causes (a cup shatters before hitting the floor).
- 5Third-party ratings reflect strong reception: G2: 4.6/5, Capterra: 4.7/5, TrustPilot: 4.5/5. Users praise the interface and output quality but criticize high costs and slow relaxed queues.
What Is Runway Gen 4.5?
Runway Gen 4.5 is the latest AI video generation model from Runway, released in December 2025. It builds on the Gen 4 architecture with significant improvements to motion realism, visual consistency, and prompt adherence. The model supports text-to-video (describe what you want and it creates it from nothing) and image-to-video (upload a still image and it animates it).
What makes Gen 4.5 stand out from its predecessor is physics. Objects in Gen 4.5 videos move with realistic weight and momentum. Pour water in a Gen 4 video and it looked like a visual effect. Pour water in Gen 4.5 and the fluid dynamics actually resemble real liquid. Fabric drapes naturally, hair moves with wind, and collisions produce believable reactions. This isn't perfect—we'll cover the failures later—but it's a genuine step forward.
The model also improved multi-element composition. When you prompt a scene with multiple characters or objects, Gen 4.5 is better at keeping each element visually consistent across the full clip. In Gen 4, a character's clothing color might shift mid-frame or background objects would morph. Gen 4.5 reduces these artifacts significantly, though it doesn't eliminate them entirely.
Quick Specs
Generation Modes:
- • Text-to-Video (describe scene + motion)
- • Image-to-Video (upload image, describe motion)
- • Video lengths: 5, 8, or 10 seconds
- • Resolution: 720p HD, 1080p Full HD (4K upscale available)
- • Aspect ratios: 16:9, 4:3, 1:1, 3:4, 9:16, 21:9
Benchmark Position:
- • #1 on Artificial Analysis Text to Video (1,247 Elo)
- • G2 rating: 4.6/5
- • Capterra rating: 4.7/5
- • TrustPilot rating: 4.5/5
- • Credit cost: 25 credits/second
Worth noting: Gen 4.5 does not include native audio generation. Your videos come out silent. Runway says sound support is coming, but there's no announced timeline. For now, you'll need to add audio in post-production using a separate tool.
How We Tested
We ran Runway Gen 4.5 on the Pro plan ($28/month) for three weeks, generating over 200 individual clips. Our goal was to test it as a practical creative tool, not just run cherry-picked demo prompts. We used it for the kinds of tasks a freelancer, small agency, or content creator would actually face.
Testing Methodology
Product and e-commerce shots (45 clips)
Animating product images, creating lifestyle contexts, pouring liquids, fabric movement on clothing
Character and storytelling scenes (60 clips)
People walking, talking, interacting with objects, multi-character scenes, facial expressions
Nature and environment footage (40 clips)
Water, fire, wind effects, landscapes, weather transitions, time-lapse style content
Social media content (35 clips)
Vertical 9:16 clips, quick transitions, text overlay bases, thumbnail animations
Stress tests and edge cases (25 clips)
Complex physics, rapid motion, multiple simultaneous actions, contradictory motion cues, maximum duration
We ran the same 15 baseline prompts through Gen 4, Gen 4.5, and (where applicable) Sora and Kling for comparison. All clips were evaluated for motion realism, prompt accuracy, visual consistency, and artifact frequency. We burned through roughly 8,500 credits across the testing period—about four months' worth of Pro plan credits, supplemented with purchased top-ups.
Step-by-Step Tutorial: Text-to-Video
Here's the exact workflow we use for generating text-to-video clips with Gen 4.5. We'll walk through the interface, prompting strategy, and settings that consistently produce the strongest results.
Step 1: Select the Model
Log into runwayml.com and navigate to the Video creation mode. In the bottom-left corner, click the model selector dropdown and choose Gen-4.5. Make sure you're not accidentally on Gen-4 or Gen-4 Turbo—they look similar in the dropdown but produce different quality and cost different credits.
Tip: Gen-4 Turbo costs only 5 credits/second compared to Gen 4.5's 25 credits/second. For initial prompt testing, consider drafting with Gen-4 Turbo first, then switching to Gen 4.5 once you've refined your prompt. This can save 80% on wasted credits during iteration.
Step 2: Write Your Prompt
For text-to-video, your prompt needs to describe both the visual scene and the motion. This is different from image generation prompts where you only describe what something looks like. Use this structure as a starting point:
Example: "Slow tracking shot of a woman in a red coat walking through a rain-soaked Tokyo street at night, neon signs reflecting in puddles, camera following from behind at shoulder height."
Keep it focused. Our testing showed that prompts under 40 words consistently outperformed longer, overly detailed ones. Gen 4.5 handles simple clarity better than dense instruction sets. Add complexity only after you confirm the base scene works.
Step 3: Use Timestamp Prompting for Sequences
When you need multiple actions to happen in order, timestamp prompting helps control the sequence. It's not frame-accurate, but it guides the general timing:
[3s] Steam rises from the cup as the camera pulls back.
[6s] A hand reaches in from the right to pick up the cup.
In our testing, timestamp prompting worked well about 70% of the time. The remaining 30%, actions shifted by 1–2 seconds or overlapped slightly. It's most reliable with 2–3 actions in a 10-second clip. More than that and the model starts dropping or merging actions.
Step 4: Configure Settings
Before hitting Generate, review these settings:
- Duration: 5s (125 credits), 8s (200 credits), or 10s (250 credits). Start with 5 seconds for testing. Use 10 seconds only when your prompt requires sequential actions or slow camera movements.
- Aspect ratio: 16:9 for cinematic content, 9:16 for social media, 1:1 for ads. We found 16:9 produces the most consistent quality.
- Resolution: 720p for drafts, 1080p for final output. 4K upscaling is available post-generation for Pro+ plans.
Credits don't roll over between billing periods. Plan your monthly generation budget accordingly—2,250 credits on Pro gives you around 90 seconds total of Gen 4.5 footage.
Step 5: Generate and Iterate
Click Generate. Processing typically takes 30–90 seconds depending on duration and server load. Review the output carefully before generating variations—each attempt costs the same credits.
If the result is close but not right, refine your prompt rather than regenerating the same one. Small wording changes ("walking briskly" vs "walking slowly", "golden hour lighting" vs "harsh midday sun") often produce dramatically different results. Our most efficient workflow was: draft prompt on Gen-4 Turbo (5 credits/sec) for 2–3 iterations, then run the refined version once on Gen 4.5.
Image-to-Video: Animating Still Images
Image-to-video was our most-used mode, and it consistently produced better results than text-to-video. The reason is straightforward: when the model doesn't have to imagine the visual scene from scratch, it can focus entirely on motion quality.
How to Use Image-to-Video
- 1Upload your image: Drag and drop it into the prompt window or click to browse. Any image works—photographs, AI-generated images, illustrations, even sketches.
- 2Write a motion-focused prompt: Do NOT describe what's in the image. The model already sees it. Describe only what should move and how. Bad: "A woman standing on a beach." Good: "The camera slowly pushes in as the woman turns her head toward the ocean, wind catching her hair."
- 3Match motion to implied physics: If your source image shows a person mid-stride, prompt for continued walking. Prompting for a completely different motion (like sitting down) requires more iterations to get right.
- 4Use the structure: "The camera [motion] as the subject [action]." Example: "The camera dollies forward as the barista pours steamed milk into the latte, creating a rosetta pattern."
One practical tip we discovered: images with implied motion (motion blur, mid-action poses, directional lines) animate more naturally than completely static compositions. A photo of someone mid-jump produces better video than a formal portrait, because the model has visual cues about the direction and nature of the movement.
Image-to-video is where Gen 4.5 really justifies its credit cost. Product photographers can turn a single hero shot into a dynamic social media clip. Illustrators can animate their artwork. E-commerce teams can create lifestyle video from product photos without hiring a videographer. The quality gap between Gen 4 and Gen 4.5 is most visible in this mode.
Pricing: What Each Plan Actually Gets You
Runway's pricing is credit-based, and Gen 4.5 is their most expensive model. Understanding the real cost per clip matters more than the monthly price tag, because credit burn rate catches a lot of people off guard.
| Feature | Free | Standard ($12/mo) | Pro ($28/mo) | Unlimited ($76/mo) |
|---|---|---|---|---|
| Monthly Credits | 125 (one-time) | 625 | 2,250 | Unlimited relaxed + 2,250 priority |
| Gen 4.5 Seconds | ~5 seconds total | ~25 seconds | ~90 seconds | Unlimited (relaxed queue) |
| Gen 4 Turbo Seconds | ~25 seconds | ~125 seconds | ~450 seconds | Unlimited (relaxed queue) |
| 4K Upscaling | No | No | Yes | Yes |
| Watermark-Free | No | No | Yes | Yes |
| Commercial Use | No | Limited | Full | Full |
| Extra Credits | No | $0.01/credit | $0.01/credit | $0.01/credit |
The real math: A single 10-second Gen 4.5 clip costs 250 credits. On the Pro plan, that's nine clips per month before you need to buy extra credits. If you're iterating on prompts (and you will be—expect 2–4 attempts per usable clip), your practical output is more like three to four finished 10-second videos per month on Pro.
Our recommendation: Start with Standard ($12/month) to learn prompting on Gen-4 Turbo, which costs only 5 credits/second. Once you're confident in your prompt style, upgrade to Pro for Gen 4.5 quality and commercial rights. The Unlimited plan at $76/month only makes sense if you're generating video daily and can tolerate the slower relaxed queue for most of your output.
Credits do not roll over. If you don't use your 2,250 Pro credits in a billing cycle, they're gone. This is one of the most common complaints in user reviews, and it's worth factoring into your decision. If you only need AI video occasionally, paying for credits on demand through the API ($0.01/credit) might be more cost-effective than a monthly plan.
Runway Gen 4.5 vs Sora vs Kling vs Pika
Gen 4.5 doesn't exist in a vacuum. Sora, Kling, and Pika are all viable alternatives with different strengths. We ran the same set of prompts through each to compare. For a deeper dive on the Sora comparison specifically, see our Sora 2 vs Runway Gen 4.5 head-to-head.
| Feature | Runway Gen 4.5 | OpenAI Sora | Kling AI | Pika 2.5 |
|---|---|---|---|---|
| Physics Realism | Excellent | Excellent | Good | Decent |
| Max Duration | 10 seconds | 20 seconds | 10 seconds | 10 seconds |
| Resolution | 1080p (4K upscale) | 1080p | 1080p | 720p |
| Lip Sync | Basic | Good | Excellent | Basic |
| Image-to-Video | Excellent | Good | Good | Good |
| Entry Price | $12/month | $20/month (via ChatGPT Plus) | $12/month | Free tier available |
| Creative Effects | Standard | Standard | Standard | Pikaffects (unique) |
| Ideal For | Cinematic, product shots | Photorealistic scenes | Talking head, dialogue | Quick social content |
Runway vs Sora: Sora produces slightly more photorealistic lighting and can generate longer clips (up to 20 seconds). But Runway's image-to-video mode is stronger, and its creative control tools (aspect ratio options, 4K upscaling) are more mature. Sora requires a ChatGPT Plus subscription, so the entry point is $20/month vs Runway's $12/month. If you're already paying for ChatGPT, Sora is essentially free to try.
Runway vs Kling: Kling wins on lip-sync and dialogue scenes. If your use case involves talking characters, Kling is currently the stronger choice. But for everything else—product shots, nature footage, cinematic sequences—Gen 4.5 produces higher quality output with fewer artifacts.
Runway vs Pika: Pika is the most accessible option with a functional free tier and lower credit costs. Its unique Pikaffects feature lets you apply creative transformations to video elements in ways none of the others can match. But raw video quality is a tier below Gen 4.5 and Sora. Pika is the right choice for high-volume social content where speed matters more than cinema-grade output. For a broader look at how autonomous AI tools are evolving beyond just video generation, see our guide to agentic AI tools.
What Runway Gen 4.5 Gets Right
Physics That Actually Convince
We generated a clip of coffee being poured into a clear glass mug. The liquid streamed naturally, hit the bottom of the mug with a realistic splash, and settled with proper surface tension. In Gen 4, the same prompt produced liquid that looked like it was being rendered in a 2010 video game. Gen 4.5's handling of fluids, fabric, and particle effects (smoke, steam, rain) is a genuine leap. It's not perfect—about 1 in 5 fluid clips had visible artifacts—but the success rate is high enough for practical use.
Image-to-Video Is a Killer Feature
The ability to take any image—a product photo, a DALL-E generation, a hand-drawn sketch—and turn it into a video clip is genuinely transformative for small creative teams. We animated product shots for an e-commerce client in about 20 minutes that would have taken a half-day video shoot. The motion quality from a good source image is consistently higher than text-to-video, and it's more predictable since the model isn't guessing at visual style.
Aspect Ratio Flexibility
Six aspect ratio options (including 9:16 portrait and 21:9 ultrawide) mean you can generate content directly formatted for Instagram Reels, YouTube, cinema-style presentations, or square ad formats without cropping or reformatting in post. This sounds minor, but it saves significant time when producing content across multiple platforms. Most competitors offer fewer ratio options.
Character Consistency Across Frames
In Gen 4, a character's hair color might shift from brown to auburn halfway through a clip, or their shirt pattern would subtly change. Gen 4.5 dramatically reduces these consistency issues. Across our 60 character-focused test clips, only 8 showed visible character mutation—compared to around 25 with Gen 4 using the same prompts. For multi-shot storytelling or any project where visual continuity matters, this improvement alone justifies the upgrade.
Where Runway Gen 4.5 Falls Short
We flagged these issues consistently across our three weeks of testing. Some are inherent to the current state of AI video generation; others are Runway-specific frustrations.
Credits Vanish Fast
At 25 credits per second, Gen 4.5 is five times more expensive than Gen-4 Turbo. The Pro plan's 2,250 credits sounds generous until you realize that's nine 10-second clips—and you'll typically need 2–4 attempts to get a usable result. In practice, Pro users can expect around 3–4 polished clips per month before buying extra credits. This is the most common criticism in TrustPilot reviews, and it's legitimate.
No Native Audio
Every video comes out completely silent. For social media content, product demos, or any video that needs sound, you'll need a separate audio tool. Runway says sound is coming but hasn't committed to a date. Competitors like Kling are further ahead here. This is a genuine workflow friction—adding sound in post adds time and sometimes a second subscription cost.
Causal Reasoning Glitches
Runway themselves acknowledge this limitation: effects sometimes precede their causes. We saw a glass shatter before it hit the floor, a ball bounce before making contact with the surface, and splash effects appear before rain started. These are intermittent (roughly 15% of physics-heavy clips) but jarring when they happen. You can't predict or prevent them through prompting—the only solution is to regenerate and hope for a better result.
Object Permanence Issues
Objects occasionally appear or disappear mid-clip without explanation. A coffee cup on a table might vanish as the camera pans, or an extra object materializes in the background. This happened in about 10% of our clips with complex scenes. Simpler compositions (single subject, clean background) are much more reliable. If your scene has more than three distinct objects, expect some permanence artifacts.
Relaxed Queue Is Painfully Slow
The Unlimited plan advertises "unlimited relaxed" generation, but TrustPilot reviewers and our own testing confirm that the relaxed queue can take 5–15 minutes per clip during peak hours. For iterative creative work where you're testing variations, this kills the feedback loop. The priority queue is much faster (under 90 seconds typically), but it uses the same limited credit pool as Standard and Pro.
None of these are deal-breakers if you go in with realistic expectations. Gen 4.5 is a massive improvement over Gen 4, but it's still AI video generation—a field that requires iteration, patience, and acceptance that not every generation will be usable.
Frequently Asked Questions
How much does Runway Gen 4.5 cost per second of video?
Gen 4.5 costs 25 credits per second. On the Pro plan ($28/month), you receive 2,250 credits—enough for roughly 90 seconds of Gen 4.5 output total. A 10-second clip costs 250 credits. Additional credits can be purchased at $0.01 each. For comparison, Gen-4 Turbo costs only 5 credits per second, and Gen-4 Standard costs 12 credits per second.
Can I use Runway Gen 4.5 for free?
Technically yes, but barely. Runway's free tier includes 125 one-time credits. Since Gen 4.5 costs 25 credits per second, free users can generate about 5 seconds of video total. That's a single short test clip. The Standard plan at $12/month (annual billing) provides 625 credits for roughly 25 seconds of Gen 4.5 per month. For serious use, the Pro plan at $28/month with 2,250 credits is the practical minimum.
What video lengths and resolutions does Gen 4.5 support?
Gen 4.5 supports durations of 5, 8, or 10 seconds per generation. Available resolutions are 720p HD and 1080p Full HD, with 4K upscaling available on Pro plans and above. Aspect ratio options include 16:9 (widescreen), 4:3 (standard), 1:1 (square), 3:4 (vertical), 9:16 (portrait), and 21:9 (ultrawide).
What is the difference between Runway Gen 4 and Gen 4.5?
Gen 4.5 improves on Gen 4 in three core areas: physics accuracy (objects move with realistic weight and momentum, fluid dynamics are convincing), visual consistency (characters and objects maintain their appearance across frames with far fewer mutations), and prompt adherence (better at following complex instructions with multiple elements). The tradeoff is cost—Gen 4.5 uses 25 credits/second compared to 12 for Gen 4 Standard and 5 for Gen 4 Turbo.
Is Runway Gen 4.5 better than Sora?
It depends on what you're creating. Gen 4.5 leads the Artificial Analysis benchmark (1,247 Elo), excels at image-to-video animation, and offers more aspect ratio options. Sora produces more photorealistic lighting effects and supports longer clips (up to 20 seconds). Runway starts at $12/month while Sora requires ChatGPT Plus at $20/month. For product animation and multi-format content creation, Runway has the edge. For long-form cinematic scenes, Sora is often the better choice.
How do I write effective prompts for Gen 4.5?
For text-to-video, use this structure: [Camera movement] shot of [subject] [action] in [environment]. Keep prompts under 40 words for best results. For image-to-video, describe only the motion since the visual elements are already in the image. Use timestamp prompting ([0s], [3s], [6s]) to control when specific actions occur in longer clips. Start simple and add detail in subsequent iterations rather than front-loading complex instructions.
Can I use Runway Gen 4.5 videos commercially?
Yes, on the Pro plan ($28/month) and above. Pro includes full commercial use rights and watermark-free exports. The Standard plan ($12/month) includes a visible Runway watermark, and the free tier is for personal experimentation only. If you need videos for client work, advertising, or any commercial purpose, the Pro plan is the minimum tier required.
Final Verdict
Runway Gen 4.5 earns its spot at the top of the AI video benchmark. The physics improvements are real and immediately visible—fluids, fabrics, and object interactions look genuinely convincing in ways that Gen 4 never achieved. Image-to-video is the standout feature that sets it apart from competitors, and the range of aspect ratios makes it practical for multi-platform content creation.
But the credit economy is punishing. At 25 credits per second, casual experimentation gets expensive fast. The lack of audio means every video needs post-production work. And the causal reasoning glitches—effects happening before their causes—are the kind of uncanny-valley artifacts that can ruin an otherwise beautiful clip.
Our recommendation: if you're a content creator, marketer, or small agency that regularly needs short video clips, Gen 4.5 is the most capable tool available right now. Start on the Standard plan to learn prompting with the cheaper Gen-4 Turbo model, then upgrade to Pro when you need commercial-quality output. If you're a hobbyist or only need video occasionally, Pika's free tier or Sora through an existing ChatGPT subscription may be more cost-effective. For a broader perspective on how AI creative tools compare to AI productivity subscriptions, our ChatGPT Plus vs Claude Pro comparison provides useful context on evaluating AI tool value.
Industry-leading quality, steep credit costs
Artificial Analysis Text to Video benchmark
~90 seconds of Gen 4.5 per month
Third-party ratings: Runway holds a 4.6/5 on G2, 4.7/5 on Capterra, and 4.5/5 on TrustPilot. Positive reviews consistently praise the interface and output quality. Negative reviews focus on credit costs, slow relaxed queues, render failures, and customer support response times. These ratings cover Runway as a platform, not Gen 4.5 specifically—the model is too new for dedicated model-level reviews.
Runway
AI-powered creative video generation
Related Articles
Sora 2 vs Runway Gen 4.5
Same prompts, two models. Which AI video generator delivers better results?
Agentic AI Tools Explained
What autonomous AI agents are, how they work, and which ones actually deliver
ChatGPT Plus vs Claude Pro
Side-by-side comparison of the two leading AI subscriptions at $20/month