Sourceful: Riverflow V2.5 Pro

Sourceful: Riverflow V2.5 Pro is an image generation model from Sourceful, and it sits at the top of the company's Riverflow 2.5 lineup. Sourceful describes it as the most powerful variant in that series, aimed at work where control and output quality matter most. The Riverflow 2.5 family is built as a unified system covering both text-to-image and image-to-image generation, and this Pro model is the version to pick when the standard tiers do not give you enough control.

The model accepts text and image inputs and returns images. Its context window is 32,768 tokens, with a maximum output of 29,491. Generation is configured through a set of image parameters: resolution can be 1K, 2K or 4K; aspect ratio can be 1:1, 4:3, 3:4, 3:2, 2:3, 16:9, 9:16, 21:9 or auto; output format can be png, jpeg or webp; and background can be auto, transparent or opaque. You can pass up to 10 input references, which is what makes image-to-image and reference-guided work possible, and n is fixed at 1, so each request produces a single image.

Those capabilities point to a fairly specific set of uses. Because it takes image references alongside text, it suits image-to-image editing, style transfer and variations on an existing asset rather than pure generation from a prompt alone. The resolution and aspect ratio options make it usable for production assets at different shapes, from square social formats to wide 21:9 compositions and vertical 9:16 ones. Transparent background output is useful when the result has to be composited into another design. Passing several references lets you steer a result with more than one source image, which helps when you need consistency across a set of outputs.

You call it through api14's OpenAI-compatible API using the model id "sourceful/riverflow-v2.5-pro". The supported parameters are include_reasoning, reasoning and reasoning_effort, alongside the image parameters described above. Because the interface follows the OpenAI format, you can point an existing OpenAI client at api14 and change the model id rather than rewriting your integration. The reasoning parameters let you control how much internal reasoning the model applies before producing an image, which is worth tuning if you find quality varies across prompts.

One practical limit to keep in mind is that n is capped at 1, so a request returns one image and batch work means multiple calls. Input references are capped at 10. If your task is text-only and does not involve images at all, an image model is the wrong tool, and a text model will fit better.