AI&Design
All change articles

Runway · Announcement · Official source

Comparing video models inside Runway

On February 20, 2026, Runway announced third-party models for paid plans. The named additions included Kling 3.0, Kling 2.6 Pro, Kling 2.5 Turbo Pro, WAN2.2 Animate, GPT-Image-1.5 and Sora 2 Pro. Runway had already added an image-to-video mode to its own Gen-4.5 in January. With both options in the app, a creator could try an in-house model and a third-party model for the same stage of production. Runway’s changelog dates Seedance 2.0 separately to April 7. The selection shown in later demonstrations therefore reflects a catalog that kept changing after the February announcement. Runway changelog

A test on Runway Unlimited

On May 16, the creator CGAlchemist published an image-to-video test conducted on Runway’s Unlimited plan. The creator’s description identifies the host and plan. The chapter list starts with a prompt, then moves through Runway Gen-4, Gen-4 Turbo and Gen-4.5, followed by examples labeled Kling 3.0 4K, Kling 3.0 Pro and several Kling O3 variants. Later sections are labeled Happy Horse 1.0 and Seedance 2.0. At 05:30 the chapter list identifies a side-by-side segment, followed at 05:55 by an upscale comparison. The uploader reports that some of the examples had sound. CGAlchemist’s May 16 comparison and chapters

That sequence documents a creator running model alternatives and reviewing the results within a single hosted workflow. Kling 3.0 connects it directly to the February addition. The visible evidence for this walkthrough is the creator’s description and chapter list. The public text record doesn’t give the source image, exact prompt, settings for each generation, or frames from the comparison. The results can establish which models the creator exercised and that they reported comparing outputs. The record leaves open which clips appeared together and whether the controls were equal across models. Its publication falls nearly three months after the announcement. CGAlchemist’s comparison · Runway changelog

What changes when the host changes

A June 16 Reddit submission reported using both Kling 3.0 and Veo 3.1 inside Runway. The submitter said they reused the Kling prompt on the direct Kling service and saw stronger color depth and sharper faces there. They described choppier motion and details blurring sooner in the Runway-hosted Veo result than in generations from similar prompts in Google AI Studio. The post includes an eight-second video with audio, but the author leaves the model and host of that attachment unspecified. The post supplies no matching pair or per-job settings. Its observations are the submitter’s reactions to their own attempts. The cause of the reported differences remains open. June 16 Reddit submission

The May test shows where the February change enters day-to-day production. A creator can try the Runway and Kling paths for an image-to-video shot, go through other options available on the account, and review the resulting takes. The June report shows a different comparison, in which one creator moved between hosts as well as models. The two accounts establish practical use of the hosted alternatives well after launch. A model-quality ranking or a host effect would need identifiable output pairs and the conditions that produced each clip. CGAlchemist’s comparison · June 16 Reddit submission