Black Forest Labs: FLUX Video Upscale

Black Forest Labs: FLUX Video Upscale is a video model from Black Forest Labs, the same provider behind the FLUX family of image models. It sits in the video category and does one job: it takes an existing video and enlarges it. The provider released it on 19 August 2026. If you already know Black Forest Labs from their image work, this is their entry into post-processing video rather than generating it from scratch.

The model accepts text and video as input and returns video as output. Its core behaviour is spatial upscaling: it enlarges a single source video by a factor between 1.5x and 3x while preserving the original duration, so the clip keeps its length and timing and only its resolution changes. An optional prompt can accompany the video input, which lets you steer the upscale with text rather than running it blind. Audio is not generated, so plan for the soundtrack to come from your source file or a separate step. No context length is published for this model, which is expected for a video pipeline where the practical constraint is the size and length of the clip you submit rather than a token budget.

The use cases follow directly from that behaviour. If you have archive footage, older renders, or client material shot at a lower resolution and you need it to hold up on a larger screen, this is the step that gets it there. It fits restoration and remastering work, preparing user-uploaded clips for higher-resolution playback, and cleaning up draft renders before delivery. Because the upscale factor is adjustable between 1.5x and 3x, you can pick a modest enlargement when you want to stay close to the source or a larger one when the target display demands it. The optional prompt is useful when you want the enlargement to favour particular detail in the frame.

You call it through api14's OpenAI-compatible API using the model id "black-forest-labs/flux-video-upscale". Because the interface follows the OpenAI schema, you can point existing client libraries at api14 and send your request in the usual way, with the video supplied as input alongside any prompt text. The model lists no supported parameters, so there are no sampling or generation knobs to tune on the request; the behaviour is set by the model itself and by the inputs you provide. Check the live price table on this page for current rates before you commit to a workload.

One thing to weigh is scope. This model upscales video, it does not generate it, so if you need a clip created from a text description you want a text-to-video model instead. It also does not produce audio, so any sound has to be handled separately. Since no supported parameters are exposed, you cannot adjust the output beyond choosing your source, your prompt and the upscale factor, and if your workflow depends on fine-grained control over encoding or frame handling, another model or a dedicated video pipeline may suit you better.