Chronological feed of everything captured from Replicate.
Uthana Inc. has deployed text-to-motion models (v1 VQ-VAE, v2 diffusion) on Replicate, converting textual descriptions into production-ready 3D animations output as FBX/GLB files compatible with Unity and Unreal. The create-character-v1 model automatically rigs any generated bipedal 3D character in under 30 seconds. These tools streamline 3D asset creation for game engines and animation pipelines.
Uthana Inc. has deployed text-to-motion models (v1 VQ-VAE, v2 diffusion) on Replicate, converting textual descriptions into production-ready 3D animations output as FBX/GLB files compatible with Unity and Unreal. The create-character-v1 model automatically rigs any generated bipedal 3D character in under 30 seconds. These tools streamline 3D asset creation for game engines and animation pipelines.
Try create-character-v1: https://replicate.com/uthana/create-character-v1
xAI's Grok Speech introduces TTS supporting 5 voices in 20 languages with expressive controls like [laugh] and <whisper> tags. The STT component handles 25 languages, provides word-level timestamps, and includes speaker diarization. Both models are immediately available for testing on Replicate.
xAI's Grok Speech models on Replicate provide TTS in 5 voices across 20 languages, supporting expressive tags like [laugh] and <whisper>. The STT model handles 25 languages with word-level timestamps and speaker diarization. Both are immediately accessible via Replicate for integration and testing.
xAI's Grok Text-to-Speech model supports 5 voices across 20 languages, including expressive tags like [laugh] and <whisper>. The Speech-to-Text model handles 25 languages with word-level timestamps and speaker diarization. Both models are available for immediate testing on Replicate.
IBM's Granite 4.1 model family unifies language, vision, speech, and guardrails capabilities into a cohesive suite deployable on Replicate. This enables developers to construct complete AI application workflows beyond isolated demos. Available models include Granite 4.1 8B for language and Granite Speech 4.1 2B for speech processing.
PrunaAI's P-video-avatar model generates cinematic-quality video avatars up to 3 minutes long at 1080p resolution for $0.025 per second. It features built-in TTS supporting over 20 languages with voice selection, dynamic backgrounds, and body/camera controls. The model is now live on Replicate and free to use all weekend.
PrunaAI's P-video-avatar model enables cinematic-quality video avatars with built-in TTS supporting 20+ languages and voice selection, dynamic backgrounds, and body/camera control. It generates up to 3-minute videos in 1080p resolution at a cost of $0.025 per second. The model is now live on Replicate and free for the weekend.
OpenAI's GPT Image 2 model is now available on Replicate, delivering exceptional photorealistic image generation. It excels in crisp text rendering and maintains strong adherence to input layouts, making it ideal for UI and design applications. Users can immediately test it via Replicate's platform.
OpenAI's GPT Image 2 model is now available on Replicate, delivering exceptional photorealistic image generation. It excels in crisp text rendering and maintains strong adherence to input layouts, making it ideal for UI and design applications. Technical users can deploy it immediately via Replicate's platform.
Replicate has released PixVerse V6, making the latest version of this AI video generation model accessible via their platform. Users can immediately test it through the provided Replicate link. This update enables rapid experimentation with enhanced video synthesis capabilities.
Replicate's X feed features an hourly poll promoting PixVerse V6, a new AI model accessible via their platform. Users can immediately test it at the provided link to generate videos from inputs. This highlights Replicate's focus on rapid deployment of cutting-edge generative AI tools.
Moonshot AI's 1T-parameter Kimi K2.6 model, now live on Replicate with open weights, autonomously refactored an 8-year-old trading engine over 13 hours. It processed 4,000+ lines of code via 1,000+ tool calls, delivering a 185% throughput gain. This demonstrates production-grade agentic coding capabilities in large-scale open models.
Replicate has released Seedance 2, a state-of-the-art text-to-video model excelling in high-fidelity motion, physics simulation, and diverse styles. It supports cinematic camera movements and complex prompts with superior temporal consistency compared to prior models. Technical users can deploy it instantly via Replicate's platform for scalable video synthesis tasks.
Replicate is running a promotional giveaway of $5 credit for Seedance 2.0, available exclusively to the first 1000 users who claim it. The offer is accessible through a specific invite link and is time-sensitive. This targets users in their X feed hourly poll context.
Replicate is running a limited-time promotion through its X feed, providing $5 credit for Seedance 2.0 to the first 1000 users who claim it. The offer is accessible via a specific invite link and emphasizes urgency with "Claim it before it's gone!" This targets rapid uptake among followers monitoring hourly polls.
Gemini 3.1 Flash TTS supports style control by embedding tags like [like dracula] directly in input text. It handles 70+ languages and accepts natural language prompts for voice modulation. This allows precise, flexible speech synthesis without complex preprocessing.
Lucy Edit 2 by DecartAI enables video editing via a single text prompt, such as converting a scene into a cyberpunk neon-lit metropolis at night with holographic signs and wet street reflections. It maintains the original video's motion while applying comprehensive stylistic changes. The model is now deployed on Replicate for public access.
Lucy Edit 2 by DecartAI transforms videos using a single text prompt, such as converting a scene into a cyberpunk neon-lit metropolis at night with holographic signs and wet street reflections. It edits the entire video while maintaining the original motion dynamics intact. The model is now deployed on Replicate for accessible use.
Replicate has added the Lucy-Edit-2 model by Decart to its platform, accessible at a dedicated URL. This deployment enables users to run the model via Replicate's infrastructure. The announcement appears in an hourly poll context on Replicate's X feed.
Anthropic's Claude Opus 4.7, their most capable model, is now hosted on Replicate. It delivers a step-change improvement in agentic coding capabilities and 3x better vision performance. The model supports a 1M token context window for extended reasoning tasks.
Seedance 2.0 now supports consistent characters by using images generated from Seedream 5 (Lite) as reference or init images. The workflow involves generating character images with Seedream 5 (Lite) while setting return_byteplus_urls=true to obtain usable URLs. These URLs are then directly inputted into Seedance 2.0 or its Fast variant for animation.
Seedance 2.0 now supports consistent characters by using images generated from Seedream 5 (Lite) as reference or init images. The workflow involves generating character images with Seedream 5 (Lite) while setting return_byteplus_urls=true to obtain usable URLs. These URLs are then directly inputted into Seedance 2.0 or its Fast variant for animation.
Cloudflare's inaugural 'Agents Week' highlights the company's strategic pivot to support the burgeoning field of agentic AI. This initiative, replacing the traditional 'Developers Week', acknowledges the profound shift in web traffic from human browsing to agent-to-agent interaction. Cloudflare aims to provide the necessary compute, storage, and security infrastructure to facilitate the development and deployment of AI agents at scale.
Ideogram AI has released 'Layerize' on Replicate, a tool designed to decompose flat graphics into structured, layered design files. The system utilizes automated font style detection (H1-small) and semantic grouping of text into containers to enable post-processing editability.
Ideogram AI has released Layerize on Replicate, a tool that converts flat graphic designs into editable, layered files. This process includes automatic detection of font styles (H1, H2, body, small) and intelligent grouping of related text elements into smart containers, streamlining the design workflow.
Google DeepMind has released Lyria 3 and Lyria 3 Pro on Replicate, enabling users to generate studio-quality music. Lyria 3 Pro specifically extends the capability to create full songs up to three minutes in length, offering granular control over musical structure through prompting for intros, verses, choruses, and bridges. This iterative development enhances AI's capacity for long-form, structured audio composition.
Google DeepMind has released Lyria 3 and Lyria 3 Pro on the Replicate platform. These models enable users to generate studio-quality music. Lyria 3 Pro offers extended song generation capabilities, allowing for tracks up to three minutes in length.
Google DeepMind has launched Lyria 3 and Lyria 3 Pro on the Replicate platform, offering AI-powered music generation. This release allows users to create structured musical pieces with distinct sections like intros, verses, and choruses. Lyria 3 Pro extends the capability to generate longer tracks, up to three minutes in length, catering to studio-quality production needs.
Replicate has released Wan 2.7 Video, a new model capable of generating, editing, cloning, restyling, and continuing video content. This model supports multimodal control inputs, including text, image, audio, and existing video. Specific functionalities include text-to-video, image-to-video, and video editing, broadening the scope of creative video manipulation on the Replicate platform.
The Wan 2.7 Video model, newly available on Replicate, enables comprehensive video manipulation including generation, editing, cloning, restyling, and continuation. This model supports control through diverse input modalities such as text, image, audio, or existing video, offering a versatile toolset for content creation and modification.
Replicate has launched Wan 2.7 Video, a new model offering advanced multi-modal capabilities for video generation and editing. This iteration allows for diverse input modalities including text, image, audio, or video to control various video manipulation tasks. Key functionalities span generation, editing, cloning, restyling, and continuation, indicating a comprehensive toolset for video content creators and developers.
Replicate has launched Wan 2.7 Video, a multimodal AI model capable of generating, editing, cloning, restyling, and continuing video content. This model supports control inputs from various modalities including text, image, audio, and existing video, providing a versatile solution for advanced video manipulation and creation tasks. The release is accompanied by demonstrations for text-to-video, image-to-video, video editing, and reference-based video generation.
Replicate has integrated Wan 2.7, a video generation model supporting text, image, audio, and video inputs. The deployment encompasses four distinct modalities: text-to-video, image-to-video, video editing, and reference-to-video generation.
Seedream 5.0 demonstrates significant advancements in image generation and editing, offering enhanced aesthetic control, sophisticated example-based transformations, and improved logical reasoning. The model exhibits precise instruction following, enabling complex compositions and intricate edits. Furthermore, it incorporates deep domain knowledge for specialized content creation and offers robust text rendering and multi-image generation capabilities.
Recraft V4 is a new suite of AI image generation models specifically engineered for design aesthetics, offering art-directed compositions and high prompt accuracy. A key innovation is its ability to produce native, editable SVG vector outputs, which is unique among current image generation models. It includes both raster and vector versions with varying resolutions and speeds, catering to diverse design and production needs.
Isaac 0.1 is a 2B-parameter, open-weight vision-language model developed by Perceptron AI for grounded perception. This model excels at OCR, object recognition, and visual reasoning, performing comparably to larger models despite its compact size. Its capabilities include explaining reasoning with visual evidence, robust OCR in challenging conditions, and understanding spatial relationships, making it suitable for real-time and edge-constrained applications like robotics and manufacturing.
FLUX.2, developed by Black Forest Labs, is an advanced image generation model with enhanced photorealism, multi-reference editing, and enterprise-grade efficiency. It offers significant improvements over its predecessor, FLUX.1, in image detail, text rendering, and prompt following. Available on Replicate, FLUX.2 caters to professional content creators, marketers, and developers requiring scalable AI visual solutions.
Nano Banana Pro demonstrates advanced capabilities beyond typical image models, showcasing built-in logic for textual interpretation and context-aware responses, and strong character consistency across varied scenarios. The model also excels in text adherence within creative designs and possesses substantial world knowledge, despite lacking real-time data integration.
Retro Diffusion's specialized pixel art models, designed for grid-aligned and limited-palette graphics, are now accessible on Replicate. These models cater to various pixel art generation needs, from fast image creation to high-quality assets, tilesets, and consistent animated sprites. Users can integrate these capabilities into their projects via Replicate's SDKs.
Replicate, a platform for AI model sharing and execution, is joining Cloudflare to enhance its infrastructure and integrate with Cloudflare's developer platform. This acquisition aims to leverage Cloudflare's robust network and developer-centric tools to scale Replicate's AI primitives, such as Cog, and develop more advanced AI abstractions. The collaboration seeks to establish a comprehensive, distributed operating system for AI, akin to existing cloud-based ecosystems but optimized for AI workloads.
Datalab's state-of-the-art document parsing and text extraction models, Marker and OCR, are now accessible on Replicate. These models offer robust performance for converting various document formats into structured data, with Marker excelling in markdown/JSON conversion with structured extraction capabilities and OCR providing multilingual text detection. Benchmarking indicates Marker's superior performance against leading OCR systems, including GPT-4o, for PDF to markdown conversion.
Google Veo 3.1 introduces significant advancements in video generation, offering new capabilities for enhanced control and creative flexibility. Key features include "Reference to Video" for combining multiple images into coherent scenes, "First and Last Frame to Video" for precise interpolation between specified start and end points, and an improved "Enhanced Image to Video" function with intelligent content understanding. These updates enable more complex narratives and consistent visual elements in generated videos.
IBM's Granite 4.0 models are a new family of open-source small language models designed for efficiency and cost-effectiveness. They leverage a hybrid architecture combining Mamba-2 and Transformers, along with Mixture-of-Experts (MoE) routing, to enable performance on consumer-grade GPUs and efficient handling of long contexts. This makes them suitable for enterprise use cases like document summarization, RAG systems, and AI agents, with the added benefit of open-source flexibility for customization and deployment.