Riverflow 2.0 Pro: #1 AI Image Model on Artificial Analysis (2026)
Sourceful Riverflow 2.0 Pro: #1 AI image model on Artificial Analysis (2026). Agentic, 4K, text rendering, $0.15 per image on OpenRouter, Runware, Replicate.
Riverflow 2.0 Pro, released by Sourceful in February 2026, is an agentic text-to-image and image editing model that ranked #1 on Artificial Analysis in both image generation and image editing, surpassing GPT Image 1.5 and FLUX.2 [max]. Priced at $0.15 per image (1K/2K) and $0.33 per 4K, it is available on OpenRouter, Runware, and Replicate with an 8,192-token prompt context window.
Riverflow 2.0 Pro is an image generation and editing model by Sourceful, released February 2026. It ranked #1 on Artificial Analysis for both text-to-image and image editing, surpassing GPT Image 1.5 and FLUX.2 [max]. Priced at $0.15 per image (1K/2K) and $0.33 for 4K, it uses an agentic self-correction pipeline for brand-grade text rendering. Available on OpenRouter, Runware, and Replicate.
Provider: Sourceful · Family: Riverflow 2.0
Context window: 8,192 tokens
Input modalities: text, image · Output: image
About Riverflow 2.0 Pro
Riverflow 2.0 Pro is an agentic image generation and editing model developed by Sourceful, a Manchester-based AI company. Released in February 2026 as the flagship tier of the Riverflow 2.0 suite, it builds on Sourceful's October 2025 Riverflow 1 model by adding text-to-image generation alongside image editing, unified through a shared agentic reasoning pipeline. The Pro tier uses an internal multi-step self-correction loop: after an initial generation pass, the model's reasoning component verifies prompt fidelity and spatial accuracy before returning the final image. This approach was explicitly designed for commercial brand design, where reliability on complex instructions matters more than raw speed. On benchmark performance, Riverflow 2.0 from Sourceful ranked #1 in both the Text to Image and Image Editing categories on the Artificial Analysis All Listings leaderboards, surpassing GPT Image 1.5 (high), Nano Banana Pro (Gemini 3 Pro Image), and FLUX.2 [max]. Artificial Analysis operates the most widely cited third-party image model leaderboard in 2026, making this a meaningful external validation. The ranking was confirmed by Artificial Analysis in February 2026 and was the first time a model from a non-major AI lab topped both image quality categories simultaneously. Riverflow 2.0 Pro accepts text prompts up to 8,192 tokens long, supports image-to-image editing with reference inputs, and generates outputs at resolutions up to 4K. Aspect ratio support extends to 21:9 for cinematic-format compositions. The Pro tier is the most capable in the Riverflow 2.0 family, prioritizing quality over speed, while the Fast and Standard Preview variants trade output quality for lower latency. The model's most distinctive capability is text rendering in images. Font control allows users to specify typefaces, and the agentic pipeline verifies that text elements in generated images are legible, correctly positioned, and matching the prompt. This directly addresses one of the most persistent weaknesses of diffusion models: generating readable, stylistically accurate text as part of a composed image. The capability makes Riverflow 2.0 Pro particularly competitive for packaging labels, advertising copy, and branded collateral. Pricing is structured per generated image rather than per token. The Pro tier costs $0.15 per image at 1K or 2K resolution and $0.33 per image at 4K. For context, 1,000 product mockups at 2K resolution would cost $150. The model is not available via a direct Sourceful API; access runs through third-party aggregators: OpenRouter (model ID: sourceful/riverflow-v2-pro), Runware, Replicate, and WaveSpeed AI. Enterprise and direct API access is available at riverflow.ai. Deployment is fully API-based with no self-hosting option. The model weights are not published or downloadable. Authentication on OpenRouter uses a standard API key; Runware and Replicate have their own account-based auth. The model is hosted by one provider at a time on OpenRouter, meaning routing redundancy requires maintaining accounts on at least two platforms. Sourceful does not publish regional availability or data residency options. On safety, Riverflow 2.0 Pro includes a configurable safety checking mode. Users can enable a single-pass safety check before generation, disable checking entirely for trusted commercial pipelines, or run checks on user-submitted prompts as needed. Sourceful has not published a system card, red-team report, or third-party safety evaluation as of June 2026. Commercial use rights are governed by Sourceful's API terms of service, and model weights are fully closed. Riverflow 2.0 Pro is strongest for teams building brand-grade commercial image workflows: creative agencies generating ad variations at scale, e-commerce teams producing product mockups from text descriptions, and packaging designers who need consistent, instruction-following output with legible text. Teams that need open weights for self-hosting, lower cost per image at volume via GPU infrastructure they own, or real-time image generation at under 3 seconds should evaluate FLUX.1 [dev] or SD3.5 as alternatives. FLUX.2 [max] remains competitive on photorealism for natural imagery, though Riverflow 2.0 Pro leads on benchmark leaderboards. Riverflow 2.5 Pro launched on June 18, 2026, adding 4K output as a standard feature, font control improvements, and custom scoring for branded style consistency. Riverflow 2.5 Pro ranked #1 across all three Design Arena categories simultaneously: Image (Elo 1408), Graphic Design (Elo 1469), and Image Editing (Elo 1377). Teams starting new projects in mid-2026 or later should evaluate Riverflow 2.5 Pro first. Riverflow 2.0 Pro remains available and is a reasonable choice for teams already integrated with its API.
Pricing
$0.15 per generated image at 1K or 2K resolution; $0.33 per 4K image. Pricing is per output image, not per token. Provider pricing on OpenRouter, Runware, and Replicate may vary slightly.
Key Features
- Agentic Self-Correction Pipeline: Multi-step internal reasoning loop verifies prompt fidelity after each generation pass before returning the final image, outperforming single-pass diffusion models on complex commercial briefs.
- Best-in-Class Text Rendering: Font control and text accuracy ranked #1 on Artificial Analysis image editing benchmark, making it the top choice for packaging labels, ad copy, and branded visual content.
- 4K Output at Full Quality: Generates images at up to 4K resolution with wide aspect ratios including 21:9, suitable for billboard, print, and high-resolution digital formats.
- Agentic Image Editing: Accepts reference images alongside natural-language edit instructions; the reasoning pipeline verifies edits match the intent before finalizing the output.
- Multi-Platform API Access: Available on OpenRouter, Runware, Replicate, and WaveSpeed AI with standard API key authentication and no proprietary SDK required.
Pros
- Topped Artificial Analysis All Listings in both text-to-image and image editing (February 2026), independently verified against GPT Image 1.5 and FLUX.2 [max].
- Agentic self-correction pipeline handles complex commercial prompts more reliably than single-pass diffusion models, particularly for multi-element product scenes.
- Font control produces legible, correctly positioned text in generated images, solving a persistent weakness of standard diffusion models for brand and packaging design.
Cons
- Closed weights, API-only: no self-hosting, no fine-tuning, no air-gapped deployment option.
- Per-image pricing ($0.15 per image) makes large-batch workflows substantially more expensive than self-hosted open-weights alternatives at volume.
- Superseded by Riverflow 2.5 Pro as of June 18, 2026, which adds higher Design Arena Elo and improved 4K features; new projects should evaluate 2.5 first.
Benchmarks
Frequently Asked Questions
What is Riverflow 2.0 Pro and who built it?
Riverflow 2.0 Pro is an agentic image generation and editing model developed by Sourceful, a Manchester, UK AI company founded in 2020 by CEO Wing Yung Chan. It launched in February 2026 as the flagship tier of the Riverflow 2.0 suite, succeeding Riverflow 1 (October 2025). The model uses a diffusion architecture combined with a proprietary reasoning pipeline that runs multi-step self-correction: after an initial generation pass, the reasoning layer verifies prompt fidelity and spatial accuracy before returning the final output. This approach was built specifically for commercial brand design, where reliability on complex instructions matters more than raw speed. Riverflow 2.0 Pro ranked #1 on Artificial Analysis All Listings for both text-to-image and image editing in February 2026, surpassing GPT Image 1.5 (high), Nano Banana Pro (Gemini 3 Pro Image), and FLUX.2 [max]. The model is available on OpenRouter, Runware, Replicate, and WaveSpeed AI.
How much does Riverflow 2.0 Pro cost per image?
Riverflow 2.0 Pro uses a per-image pricing model, not per-token. Generating a 1K or 2K resolution image costs $0.15 per output image. Generating a 4K resolution image costs $0.33 per image. There is no monthly subscription or free tier. For scale: 1,000 product mockups at 2K costs $150; 20 high-resolution 4K ad creatives costs $6.60; a daily design workflow of 200 images at 2K costs $30. Pricing may vary slightly across platforms: OpenRouter, Runware, Replicate, and WaveSpeed AI each set their own rates, though most follow the reference pricing closely. Sourceful does not offer a batch discount or cached-prompt pricing as of June 2026. For very high-volume production at 10K or more images per month, self-hosted open-weights alternatives like FLUX.1 [dev] on owned GPU infrastructure are significantly cheaper.
What image resolutions and aspect ratios does Riverflow 2.0 Pro support?
Riverflow 2.0 Pro supports image generation at up to 4K resolution. The 1K and 2K output tiers are priced at $0.15 per image; 4K output costs $0.33 per image. Aspect ratio support extends to 21:9 for cinematic-format compositions, as well as standard ratios including 1:1, 16:9, 4:3, and 9:16 for portrait and social media formats. The model accepts text prompts up to 8,192 tokens, giving ample room for detailed scene descriptions. Image-to-image editing mode accepts reference images in common formats. The Pro tier generates the highest-quality output in the Riverflow 2.0 family; the Fast and Standard Preview variants offer lower latency at the cost of some output fidelity. Riverflow 2.5 Pro (June 2026) extended 4K output with additional font control improvements, so teams specifically needing 4K may also want to compare the two versions.
How does Riverflow 2.0 Pro compare to FLUX.2 [max] and GPT Image 1.5?
On the Artificial Analysis All Listings image leaderboards (February 2026), Riverflow 2.0 Pro ranked #1 in both text-to-image and image editing, placing ahead of GPT Image 1.5 (high) and FLUX.2 [max]. This was a third-party independent ranking, not vendor-reported. On text rendering specifically, the Pro tier's font control system outperforms both competitors on brand and packaging design tasks where legible, positioned text is critical. FLUX.2 [max] remains highly competitive on photorealistic natural imagery and is available in open-weights format, which makes it the better choice for teams needing self-hosting. GPT Image 1.5 benefits from OpenAI's infrastructure and tighter integration with GPT-4o for multi-modal workflows involving both text generation and image creation. Riverflow 2.0 Pro does not have a text generation capability, so workflows requiring image creation plus copy writing must pair it with a separate language model. For pure image quality on commercial design tasks, the Artificial Analysis data favors Riverflow 2.0 Pro over both competitors at the time of its launch.
Is Riverflow 2.0 Pro open source or proprietary?
Riverflow 2.0 Pro is fully proprietary. The model weights are not publicly available, not downloadable from Hugging Face or any other repository, and there is no self-hosting option. Access is exclusively through third-party API aggregators: OpenRouter (model ID: sourceful/riverflow-v2-pro), Runware, Replicate, and WaveSpeed AI, plus enterprise and direct API access at riverflow.ai. Sourceful has not published the model's architecture details, parameter count, or training data composition. The model is not licensed under any open-source or open-weights license. Commercial use rights are governed by Sourceful's API terms of service as accessed through each platform. Teams requiring open weights for compliance, air-gapped deployment, or fine-tuning should use FLUX.1 [dev] (Apache 2.0) or SDXL (modified CreativeML Open RAIL+M license) instead.
What types of image generation and editing can Riverflow 2.0 Pro do?
Riverflow 2.0 Pro supports two main modes: text-to-image generation and image-to-image editing. Text-to-image generates a new image from a natural-language prompt, accepting prompts up to 8,192 tokens. Image-to-image editing accepts a reference image plus an instruction prompt; the agentic reasoning pipeline verifies that the edit matches the instruction before returning the result. The model does not produce video, audio, or text as output. It does not support function calling, web browsing, or code execution. The Pro tier's distinguishing feature is font control for text rendering: users can specify typefaces by name and the pipeline generates legible, positioned text within the image, a capability that most diffusion models lack. Output formats include common raster image types at resolutions up to 4K and aspect ratios up to 21:9. It does not support real-time streaming generation; outputs are returned as complete images after the full generation pipeline completes.
Does Riverflow 2.0 Pro train on user data?
Sourceful states it does not train on API inputs. However, a formal data retention policy has not been published as of June 2026. Data handling terms for most users are governed by the terms of service of the API platform through which access is obtained: OpenRouter, Runware, Replicate, and WaveSpeed AI each have their own data and privacy policies. Direct enterprise access through riverflow.ai may offer different data handling agreements. Sourceful has not published SOC 2 Type II, ISO 27001, or HIPAA-eligible compliance certifications as of June 2026, which limits its suitability for enterprise environments with strict data governance requirements. Riverflow 2.0 Pro has not been classified under the EU AI Act as of this review. Teams with GDPR or regulated-industry data requirements should review the individual platform terms for the API provider they use and confirm with Sourceful directly about enterprise data handling options.
Who is Riverflow 2.0 Pro best for and who should avoid it?
Riverflow 2.0 Pro is best for brand and packaging design teams that need reliable instruction following on complex visual briefs, particularly where legible text must appear within generated images. Creative agencies generating ad variations at scale benefit from its agentic quality assurance and 4K output. E-commerce teams building product mockup pipelines from text descriptions get consistent, commercial-grade output without requiring a manual quality step for most briefs. The per-image pricing ($0.15 to $0.33) is acceptable for low-to-mid volume production but becomes expensive above roughly 10,000 images per month. Teams that should avoid Riverflow 2.0 Pro include those needing self-hosted or air-gapped image generation (closed weights, API-only), those with very high batch volumes where the per-image cost is prohibitive compared to owned GPU infrastructure running FLUX.1 [dev], and teams specifically needing video or audio generation (not supported). Teams starting new projects in mid-2026 should also evaluate Riverflow 2.5 Pro, which supersedes this model with higher Design Arena Elo rankings and improved 4K support.