What do you like best about Runway?
The reason Runway became the default in my studio is not any single model. It is that the whole thing is built like a production environment rather than a prompt box with a render button, and that difference shows up in every project we push through it.
Image-to-video with Gen-4 is where most of our actual output comes from. We rarely generate from a bare text prompt. The workflow that works is locking the look in a still first, either a photo from a shoot or a generated frame, and then letting Gen-4 handle the motion. The subject consistency is the part that changed what we could sell to clients. A character or a product that keeps the same face, the same materials, and the same lighting logic across separate generations means we can build a sequence of shots that reads as one piece, not a collage of AI clips that happen to share a color grade. Earlier generations of this technology fell apart exactly here, and Gen-4 mostly does not.
Aleph is the feature I did not expect to rely on and now use weekly. It edits footage we already have instead of generating from nothing. Bring in a real clip, describe the change, and it returns the same shot with the change applied while keeping the original motion. We have used it to swap a daytime product shot into an evening mood for a seasonal campaign, and to remove a background element a client objected to after the shoot was already wrapped. The alternative in both cases was a reshoot or an afternoon of manual VFX work. Neither happened.
Act-Two deserves its own mention because it replaces an entire category of vendor. Record a performance on a phone, upload it as the driving video, and the character in your generated footage picks up the gesture and the facial expression. It is not flawless, hands and fast movement still need a second attempt sometimes, but the fact that a two-person team can do performance transfer at all, without a mocap studio, without suits, without calibration, is the kind of capability that used to be a budget line, not a feature.
The multi-model setup quietly became one of the biggest reasons to keep the subscription. One dashboard now covers:
- Runway's own Gen-4.5, Gen-4, and Gen-4 Turbo
- Google's Veo models
- Kling and Seedance for the shots where their motion style fits better
- The image models we use for start frames
Before this, we held separate accounts across three vendors, compared outputs by downloading and re-uploading, and paid three invoices. Now the comparison happens in one interface with one asset library underneath it, and picking the right model per shot became a two-minute decision instead of a context switch.
The built-in editor closes the loop. Generations land in a timeline where we can trim, sequence, and export up to 4K without round-tripping into Premiere for drafts. For final client delivery we still finish in a proper NLE, no illusions there. But for the internal review stage, where the question is whether the sequence works at all, staying inside one tool removes a whole export-import cycle per iteration.
Explore Mode on the top plan changed how the team experiments. Unlimited generations at a relaxed rate means junior people can iterate on a concept for an hour without anyone watching a credit balance drain. That psychological shift matters more than it sounds. When every attempt costs money, people stop attempting, and the quality of the eventual output drops with it.
Commercial rights are clean. We own what we generate, on every plan, no attribution required. For an agency context this is not a footnote, it is the first question legal asks, and Runway's answer is short.
Workflows is newer and still finding its shape, but the idea is right. You chain steps, an image generation feeding into an image-to-video pass feeding into an upscale, and run the chain as one unit instead of babysitting each stage by hand. For repeatable formats, we produce a recurring social series with a fixed visual recipe, setting the recipe up once and feeding it new inputs is exactly the kind of boring automation that saves an hour a week without anyone noticing.
The lip sync and voice tools on the Pro tier get qualified approval from me. Custom voice creation paired with lip sync is good enough for internal drafts and for scratch narration a client reviews before we book a voice actor. It is not good enough to replace the voice actor on final delivery, at least not for the languages we work in, and I would not claim otherwise. As a previz tool it earns its place. As a finishing tool, not yet.
Upscaling and watermark removal on paid plans work as stated. Nothing to add, they do the job. Review collected by and hosted on G2.com.
What do you dislike about Runway?
The credit system is the part of Runway I have a genuinely complicated relationship with, and it is the first thing I warn new team members about. Credits are not a flat currency. Different models burn at very different rates, a second of one model costs a multiple of a second of another, and a generation that misses the prompt costs exactly the same as one that lands. There is no preview step before credits are committed, so a complex prompt that the model misreads is money spent on footage nobody will use. Our mitigation is a house rule: draft on Gen-4 Turbo at low cost until the composition and motion are right, and only then re-run the locked prompt on the expensive model. That discipline cut our wasted spend substantially, but the fact that we needed a house rule at all says something about how the pricing is structured.
Prompt adherence on complicated instructions is the second recurring friction. Single-subject, single-action prompts land reliably. Ask for two characters doing different things in the same shot, or a specific camera move combined with a specific object interaction, and the hit rate drops noticeably. The model tends to pick the part of the prompt it understood best and quietly ignores the rest. We work around it by splitting complex shots into simpler generations and compositing, which works, but it is planning overhead the marketing does not mention.
The relaxed queue on Explore Mode is slow when it matters most. Off-peak it is fine. On a deadline afternoon a single generation can sit for long enough that the unlimited label starts to feel theoretical, and there is no way to stack a batch of prompts and walk away, you wait for each one. For deadline work we fall back to credited generations at priority speed, which is the intended design, I suspect, but it stings on a plan at this price.
Content moderation produces false positives more often than it should. Perfectly ordinary commercial footage, people in normal summer clothing, a gym scene for a fitness client, occasionally gets flagged with no explanation of which element triggered it. The generation fails, the reason stays opaque, and you rewrite the prompt by guesswork. It has never blocked a project outright, but the trial and error wastes time and, before we adopted the draft-cheap-first rule, wasted credits too.
Vertical short-form is workable but clearly not the native format. The platform thinks in cinematic horizontal terms, and producing 9:16 content for Reels and TikTok involves more cropping and reframing than a social-first tool would require. For teams whose entire output is vertical social, this is worth weighing before committing. Review collected by and hosted on G2.com.