Skip to content
Rio-3.5 trends on Hugging Face, BiRefNet video tools hit Replicate, Anthropic industry updates | Daily AI roundup cover

Rio-3.5 trends on Hugging Face, BiRefNet video tools hit Replicate, Anthropic industry updates

By Harsh Desai
Share

TL;DR

Fresh open models appeared on Hugging Face while Replicate added background removal options for video and images. Vercel and Anthropic released policy and integration changes that affect access and workflows.

What shipped

On 13 June 2026 several model hubs and vendors shipped updates. Hugging Face and Replicate surfaced new image-text models and background tools. Anthropic and Vercel adjusted access rules and added workflow features.

Hugging Face trending

Two large image-text-to-text models rose in visibility on the Hugging Face Hub. Prefeitura-rio and DavidAU each placed a model that supports download and fine-tuning. These releases give builders ready checkpoints for image and text tasks.

  • Rio-3.5-Open-397B Prefeitura-rio placed an image-text-to-text model on the Hugging Face Hub that supports fine-tuning and inference. Teams can test it for captioning or visual question tasks without new infrastructure.
  • Qwen3.6-40B variant DavidAU released a large image-text-to-text model that is now trending on the Hub. Users can download the GGUF version for local experiments or custom code tasks.

Replicate new models

Sprited added two BiRefNet variants for background removal on Replicate. One handles video frame by frame while the other processes still images. A third model from Tattzy25 offers style controls through prompt inputs.

  • birefnet-video Sprited released a video background removal model on Replicate that outputs transparent WebM files by default. It processes each frame and supports commercial use under an MIT license.
  • heavy_hand_dark Tattzy25 added a model on Replicate that accepts prompts and LoRA scale settings for image generation. It runs through the platform API with low run counts so far.
  • birefnet Sprited launched the full set of ten BiRefNet variants plus ToonOut on Replicate. The tool returns cutouts or masks and works for both product photos and illustrations.

Vendor launches

Vercel raised storage limits and added workflow features while Anthropic issued statements on model access. The changes affect hobby users and regulated industry partnerships.

  • Claude Fable 5 suspension Vercel removed access to Claude Fable 5 on its AI Gateway after a government directive. Other Anthropic models remain available through the same gateway.
  • Anthropic access statement Anthropic published a note on the US government order to suspend Fable 5 and Mythos 5. The company confirmed other models stay accessible.
  • TCS Anthropic partnership Anthropic announced a collaboration with TCS to deliver Claude models to regulated sectors. The partnership targets compliance-heavy industries that need audited AI deployments.

Product Hunt picks

CakewordAI: The app lets users point a camera at objects to receive names in any chosen language. It targets travelers and language learners who want instant vocabulary.

Other

AWS posted several guides on agentic workflows and document pipelines. Manus added a Canva connector for design tasks.

  • Meeting prep assistant AWS showed how to connect Amazon Quick and Cisco Webex MCP servers to build a meeting preparation agent. The example pulls context before calls and drafts follow-ups afterward.
  • Rocket Close Supercharger AWS described an agentic system Rocket Close uses to speed title operations. The agents handle document checks that previously required manual review.
  • PDF document pipeline AWS outlined a generative AI flow that turns PDF stacks into structured insights. The pipeline uses multiple services to extract and summarize content at scale.
  • AWS Professional Services team AWS shared how its internal services group adopted frontier models first. The account covers training programs and production benchmarks.
  • Manus Canva connector Manus released a direct link to Canva that turns its agent into a design helper. Users can generate and edit graphics without leaving the Manus workspace.

Industry news

KPMG report withdrawal: KPMG removed a study on AI usage after reviewers found apparent hallucinations in the findings. The incident again highlighted risks when models generate statistics without verification.

What this means for you

For Vibe Builders: You can call the new BiRefNet models on Replicate through simple API requests to remove backgrounds from photos or video frames. The Hugging Face checkpoints give you starting points for image tasks you can fine-tune in the browser. The AWS guides show ready patterns for meeting assistants and document flows that you can adapt without writing server code.

For Non-techies: Background removal tools now run on Replicate so you can clean product images or meeting recordings without extra software. AWS examples show how agents can prepare meeting notes and pull insights from PDFs you already store. The Canva link from Manus lets you generate design assets from text prompts inside one workspace.

For Developers: Replicate and Hugging Face now host ready checkpoints for background removal and image-text work that you can benchmark against your current stack. Vercel updated Nitro and Blob limits so hobby and production apps can scale storage without plan changes. Watch the Anthropic access changes and the AWS agent patterns for reliability signals before adding new endpoints to live systems.

What to watch next

Track whether Anthropic restores Fable 5 access and whether more teams adopt the BiRefNet video endpoint on Replicate. Look for follow-up posts from AWS on the document pipeline and meeting agent examples.

Harshs take

The day delivered mostly incremental model drops and repeated vendor posts rather than new capabilities. Duplicate announcements from Anthropic and AWS diluted signal while the KPMG withdrawal reminded everyone that model outputs still need human checks. Builders should test one Replicate background endpoint this week against their current image workflow and record the exact latency and cost before adding it to production.

by Harsh Desai

More AI news

Everything AI. One email.
Every Monday.

New tools. Model launches. Plugins. Repos. Tactics. The moves the sharpest builders are making right now, before everyone else.

No spam. Unsubscribe anytime.