Sourceful: Riverflow V2.5 Fast

Sourceful: Riverflow V2.5 Fast is an image generation model from Sourceful, the speed-optimized variant of the company's Riverflow 2.5 lineup. The Riverflow 2.5 series is a unified text-to-image and image-to-image family, and this Fast version sits inside that family as the option built for production deployments and latency-critical workflows.

The model accepts text and image inputs and produces images as output. It supports a context length of 32,768 tokens and a maximum output of 29,491 tokens. Generation is controlled through a defined set of image parameters. Resolution is an enum of 1K or 2K. Aspect ratio is an enum covering 1:1, 4:3, 3:4, 3:2, 2:3, 16:9, 9:16, 21:9 and auto. Output format is an enum with jpeg as the only value. Background is an enum of auto, transparent or opaque. The n parameter has a range of 1 to 1, so each request returns a single image. Input references have a range of 0 to 4, which is what enables image-to-image work alongside pure text-to-image generation.

Those capabilities point to a fairly clear set of uses. Prompt-only generation covers the everyday cases where a developer needs an image from a description, picking a resolution and aspect ratio to match the target layout. The transparent background option suits assets that need to be composited over other content, such as product cutouts or interface elements. The input references parameter, up to four images, supports image-to-image workflows where existing visuals guide the result, whether that means restyling, variations or edits anchored to reference material. Because this is the Fast variant, it fits pipelines where throughput and response time matter more than maximum refinement, such as generating images on demand inside an application or processing requests at volume.

You call the model through api14's OpenAI-compatible API using the model id "sourceful/riverflow-v2.5-fast". The supported request parameters are include_reasoning, reasoning and reasoning_effort, alongside the image parameters described above: resolution, aspect_ratio, output_format, background, n and input_references. If your client already speaks the OpenAI format, switching to this model is a matter of changing the model id and adding the image settings you need.

The constraints to keep in mind are concrete rather than vague. Output is jpeg only, so formats like png or webp are not available here. Each request returns exactly one image, so batches require separate calls. Input references top out at four images. If your work depends on a different output format, multiple images per call, or a model outside the image category entirely, another model will fit better.