Sourceful ResearchIntroducing Sourceful Riverflow 1, our new state-of-the-art image editing model for higher reliability editing
- Sourceful Research
- Developers
- Designers
- Image Editing
- API
Wing ChanOctober 14, 2025 - 7 min read
Today, we’re excited to introduce Sourceful Riverflow 1, a new model purpose-built for image editing. Riverflow 1 achieves state-of-the-art performance according to Artificial Analysis’s leading image editing benchmark, across a diverse set of tasks.
Riverflow was born from Sourceful’s need to push the frontier of image editing for our core use case: helping users design branded packaging and marketing collateral. In 2025, there will be an estimated 8.8 trillion pieces of branded packaging, marketing and merch printed, across nearly 390 million unique designs. Unlike cartoons, memes, social media ads and photos of people, designs destined for physical print demand tighter tolerances where millimetre-level accuracy matters. Small errors cost time and money. While image editing models are far from this goal today, every incremental improvement in control and consistency helps users visualise production-ready outcomes with confidence.
Sourceful serves an almost even split of designers and non-designers. Teams rely on AI to bring brand ideas to life faster, iterate more efficiently, and visualise an immersive brand world — capacity that was historically reserved for the largest brands with bespoke design budgets.
Riverflow 1 achieves state-of-the-art performance on Artificial Analysis's image editing benchmark.
Product Overview
Riverflow 1 is available now on the Sourceful platform and via API through launch partner Runware, with additional inference platforms coming soon. Developers can choose from three tiers calibrated for different accuracy and throughput requirements:
riverflow-1-mini— $0.05 per imageriverflow-1— $0.066 per imageriverflow-1-pro— Private beta
Each successive tier increases reasoning depth to unlock higher accuracy and consistency.
Benchmark Highlights
Riverflow 1 consistently delivers better instruction-following and higher-resolution rendering than Gemini 2.5 Flash (Nano-Banana), the prior state of the art. The following examples highlight reliability gains across common editing tasks.
Riverflow can remove imperfections like drops or scratches without changing other parts of the image.
Pushing the Frontier
By observing countless hours of artworkers inside image and vector editing tools, Sourceful’s research team identified that professionals take a multi-step approach to editing. They isolate elements, adjust colours and styles, and iterate through successive refinements. This step-by-step, reasoning-driven approach, proven in large language and video models through “chain of thought” and “chain of frames” methodologies, translates directly to image editing. Riverflow 1 packages this workflow into a controllable, reliable inference process.
Riverflow and Runware
Riverflow 1 launches day one on Runware Inference Cloud (API and Playground) and is now a core model inside Sourceful’s AI platform. Runware offers leading image, audio, and video models — proprietary and open weight — through a single interface. Their Sonic Inference Engine enables these models to run at a fraction of conventional deployment costs.
Sourceful was one of our first users, and it’s been exciting to support their growth journey as they scale with Runware. Their team has fully harnessed the power of our Sonic Inference Engine, pushing what’s possible with real-time visual editing. We’re proud to be the launch partner for Riverflow 1 and look forward to what comes next.
Ioanna Hreninciuc, Co-Founder, Runware
Developers can request Riverflow 1 on Runware using air code sourceful:1@0 (mini) or sourceful:1@1 (standard). Cloud platforms interested in hosting Riverflow 1 can reach Sourceful on LinkedIn or X.
Future Models
Riverflow is on a rapid release cadence, with new versions planned every three weeks. Riverflow 1 Pro remains in private beta; teams eager to test it can join the waiting list.
Sourceful is also developing a Brand World model that ingests brand assets — logos, colour systems, references — to generate brand-realistic imagery suited for mockups and marketing moments. Interested brands, applications, and clouds can express interest here.
Key Commercials
All models are 1K resolution, 11 Aspect Ratios and support Transparent outputs.
- riverflow-1-mini: $0.05 per image
- riverflow-1: $0.066 per image
- riverflow-1-pro TBD
One step closer to making “Create Something Iconic Today” a reality for anyone with the ambition to build.
Best,
Wing Chan
CEO, Sourceful
Manchester, United Kingdom
