AI Model Library

A mixture-of-experts model with 1 trillion parameters and 32 billion activated for agentic intelligence, tool use, and autonomous reasoning tasks.

A 2-billion-parameter multimodal transformer supporting text, images, audio, and video with memory-efficient architecture optimized for resource-constrained devices.

Multimodal text generation model with 4 billion effective parameters, optimized for mobile deployment using MatFormer architecture and elastic inference capabilities.

A 24-billion parameter multimodal model featuring improved instruction following, function calling, and reduced repetition over its predecessor.

A 70.6 billion parameter Llama 3.3-based model fine-tuned for character consistency and dynamic dialogue in creative text generation applications.

A 671B-parameter MoE model with 37B active parameters featuring enhanced reasoning capabilities through reinforcement learning and chain-of-thought training methodologies.

A 23.6B parameter coding assistant finetuned for agentic software engineering tasks with 128K context window and 46.8% SWE-Bench performance.

A 24-billion parameter language model fine-tuned on Mistral Small 3.1 Instruct, specialized for interactive storytelling and text-based adventures.

Instruction-based image editing model that modifies existing images through natural language commands while preserving unintended regions using spatially weighted loss functions.

A 0.6B parameter language model featuring dual thinking modes, multilingual capabilities, and 32K context length through strong-to-weak distillation training.

A 1.7 billion parameter multilingual transformer supporting dual-mode reasoning with step-by-step "thinking" and rapid "non-thinking" response capabilities.

A 4-billion parameter transformer model featuring dual reasoning modes, extensive multilingual training, and competitive performance across mathematical, coding, and logical reasoning benchmarks.

Dense 8.2 billion parameter transformer model featuring hybrid thinking capabilities with 32K token context and multilingual support across 119 languages.

A 14.8 billion parameter transformer model featuring hybrid thinking/non-thinking reasoning modes, 32K context length, and multilingual capabilities across 119 languages.

A 32.8 billion parameter language model featuring hybrid thinking modes for both rapid responses and step-by-step reasoning across multilingual tasks.

Mixture-of-experts model with 30.5B total parameters, 3.3B activated per token, featuring hybrid reasoning modes and multilingual support across 119 languages.

Large language model with Mixture-of-Experts architecture featuring dual operational modes for both rapid inference and complex reasoning tasks.

A 1.5B parameter code generation model fine-tuned with reinforcement learning and iterative context lengthening for long-context reasoning tasks.

A 14-billion parameter code reasoning model fine-tuned using distributed reinforcement learning with long-context capabilities up to 64,000 tokens.

17-billion parameter image generation model featuring sparse Diffusion Transformer architecture with Mixture-of-Experts and hybrid text encoding for high-quality synthesis.

A 14.8 billion parameter instruction-tuned language model trained using Iterated Distillation and Amplification with hybrid reasoning capabilities across 30+ languages.

A 32-billion parameter instruction-tuned model based on Qwen2.5 architecture featuring dual operational modes and iterated distillation alignment methodology.

A Llama 3.1-based model trained with Iterated Distillation and Amplification, featuring dual reasoning modes and tool calling capabilities.

A 70B parameter instruction-tuned model based on Llama 3.1 architecture featuring dual reasoning modes and multilingual tool-calling capabilities.

A 3B-parameter multilingual instruction-tuned model based on Llama 3.2 that supports tool-calling and features dual operational modes for standard and extended reasoning.

Large-scale MoE language model utilizing 671B parameters with 37B activated per token, featuring enhanced reasoning and multilingual capabilities.

A 24-billion parameter multimodal transformer supporting text and vision tasks with 128K token context length under Apache 2.0 license.

A 1 billion parameter multimodal transformer model trained on 2 trillion tokens with 32,000 token context window supporting text and image processing.

A 4.3 billion parameter multimodal transformer model that processes text and images, supports 128K token contexts, and operates across 140+ languages.

A 12-billion parameter multimodal transformer that processes text and images, supporting 140+ languages with 128,000-token context length.

A 27-billion parameter multimodal model processing text and image inputs with 128K context length and SigLIP vision encoder integration.

A decoder-only transformer with 256K context window, featuring multilingual capabilities, tool use, and code generation optimized for enterprise applications.

A 32.5-billion parameter transformer model optimized for mathematical reasoning, coding, and complex problem-solving through reinforcement learning training.

A 1.3B parameter diffusion transformer model for text-to-video generation supporting multilingual prompts and consumer GPU deployment.

A 14 billion parameter diffusion transformer model that converts single images into temporally coherent video sequences at 480P resolution.

A 14-billion parameter diffusion transformer model that generates 720P videos from single input images using a unified architecture with 3D causal VAE compression.

A 14-billion parameter diffusion transformer model that generates videos from text prompts, supporting both Chinese and English languages at 480P and 720P resolutions.

A 70.6-billion parameter language model fine-tuned for adventure role-play scenarios, emphasizing conflict, tension, and narrative stakes in second-person storytelling.

A fine-tuned 23.6 billion parameter Mistral-based model designed for long-context conversations and maintaining narrative coherence across extended dialogues.

A 24-billion parameter instruction-tuned model built on Mistral architecture with deliberately removed content filters to maximize user control over outputs.

A 3B-parameter multimodal language model that processes images, videos, and text with capabilities for visual reasoning, document analysis, and computer interface interaction.

A 7-billion parameter multimodal model capable of processing images, documents, and videos with native dynamic resolution support and structured output capabilities.

A 72-billion parameter multimodal model that processes images, videos, and text for document analysis, object detection, and visual agent tasks.

A 2 billion parameter flow-based diffusion transformer that generates high-resolution images from text descriptions using efficient attention mechanisms.

A 1.5 billion parameter language model created through distillation techniques, focusing on mathematical reasoning and chain-of-thought problem-solving capabilities.

A 7.62B parameter distilled language model based on Qwen2.5-Math-7B, trained via knowledge distillation for mathematical and logical reasoning tasks.

A 14B dense language model distilled from a mixture-of-experts architecture, optimized for mathematical reasoning and code generation tasks.

A 32B parameter language model created through knowledge distillation, optimized for mathematical reasoning, code generation, and complex problem-solving tasks.

Distilled 8B-parameter model optimized for mathematical reasoning and code generation through knowledge transfer from larger reinforcement learning-trained teacher models.

A 70B parameter dense language model distilled from DeepSeek-R1 using Llama 3.3 architecture, optimized for mathematical and coding reasoning tasks.

A 671B parameter Mixture-of-Experts model trained with reinforcement learning to enhance reasoning capabilities in mathematics, coding, and logical tasks.

A fine-tuned 8-billion parameter text-to-image model using rectified flow transformer architecture with an optional guidance embedder for CFG-free generation.

Large language model with 456 billion parameters featuring hybrid attention mechanism and ultra-long context support up to 4 million tokens.

A 7B parameter process reward model that evaluates mathematical reasoning steps individually rather than just final answers, achieving 67.6% accuracy on mathematical benchmarks.

A 72.8 billion parameter process reward model that evaluates intermediate mathematical reasoning steps using consensus filtering and achieves 78.3% F1 on ProcessBench.

A 7-billion parameter transformer model with 1-million token context capacity utilizing Dual Chunk Attention for long-range text processing.

A 14.7B parameter transformer model with 1-million token context capacity, utilizing dual chunk attention and progressive training for extended sequence processing.

A 70.6-billion parameter language model finetuned from Llama 3.3 using extended training and checkpoint merging techniques for improved output stability.

An uncensored instruct-tuned 3.2B parameter language model that grants users full control over system prompts and behavioral alignment.

A 671-billion parameter Mixture-of-Experts language model with 37 billion active parameters per token, featuring auxiliary-loss-free load balancing and FP8 mixed-precision training.

A Flux.1-based checkpoint model fine-tuned with an expanded dataset for generating highly realistic photographic imagery with enhanced anatomical accuracy and texture fidelity.

A Mixture-of-Experts vision-language model series featuring dynamic image tiling, visual grounding capabilities, and efficient sparse computation across three parameter variants.

A 2.8B parameter mixture-of-experts vision-language model with dynamic tiling for multimodal understanding, OCR, and visual grounding tasks.

A compact Mixture-of-Experts vision-language model with 1.0B activated parameters supporting multimodal tasks including OCR, document analysis, and visual grounding.

A 14-billion parameter decoder-only transformer optimized for mathematical reasoning, code generation, and scientific tasks through curated synthetic training data.

A 70.6-billion parameter text generation model fine-tuned from Llama 3.3, designed for creative writing and role-playing applications.

A 70-billion parameter language model fine-tuned from Llama 3.3 for creative writing and role-playing applications using custom datasets.

Llama 3.3 is a text-only 70B instruction-tuned model that provides enhanced performance relative to Llama 3.1 70B and to Llama 3.2 90B when used for text-only applications. For some applications, Llama 3.3 70B approaches the performance of Llama 3.1 405B.

Open-source text-to-video and image-to-video generation model using a 13-billion parameter Transformer architecture with 3D variational autoencoder.

A 123-billion parameter language model optimized for conversational AI, creative prose generation, and role-playing applications with enhanced narrative consistency.

Text-to-video diffusion transformer with 3D VAE compression generating 1360x768 resolution clips up to 10 seconds duration.

A 5-billion parameter diffusion transformer model that generates temporally coherent videos from input images and text prompts.

Experimental research model focused on advancing AI reasoning capabilities by thinking through problems step by step, continuously questioning assumptions, and exploring different paths of thought. Impressive benchmark scores in math and coding tasks.

Qwen2.5-Coder builds on Qwen2.5 by training on 5.5 trillion additional tokens of code data including source code, text-code grounding data, and synthetic data. This leads to significant improvements in code-related tasks.

This 8-billion parameter model brings improvements to SD3 in customizability, efficiency, and diversity. This Large variant was designed for professional use cases at 1 megapixel resolution.

This latest iteration in the Stable diffusion family brings improvements over SD3 in customizability, efficient performance, and diverse outputs. The Turbo variant is distilled from the Large version to generate images in just 4 steps.

The next iteration in the Llama series of open models. This lightweight model was designed to run on edge devices, even mobile.

A 5-billion parameter transformer-based diffusion model that generates short video sequences from input images and text prompts.

A 1.5 billion parameter specialized model focused on mathematical reasoning and problem-solving capabilities in English and Chinese.

A 7.62-billion parameter mathematical reasoning model supporting English and Chinese with chain-of-thought and tool-integrated reasoning capabilities.

A 72.7 billion parameter language model specialized for solving mathematical problems in English and Chinese using chain-of-thought and tool-integrated reasoning.

A 7.61 billion parameter transformer model designed for code generation and reasoning across 92 programming languages with a 128,000-token context window.

A 7.61-billion parameter transformer-based language model with 128K token context length, trained on 18 trillion tokens supporting 29+ languages.

A 14.7 billion parameter transformer model supporting 29 languages with 128K context window, designed as foundation for fine-tuning applications.

A 32.5 billion parameter multilingual transformer model trained on 18 trillion tokens with 128K context length supporting text generation, coding, and mathematical reasoning.

This update to Qwen 2, pretrained on 18 trillion tokens, shows impressive performance in benchmarks (MMLU 85+, HumanEval 85+, MATH 80+), outperforming models of similar size. Qwen 2.5 has multilingual support for over 29 languages and can understand and generate structured data.

A 22B parameter enterprise-grade small model, a convenient mid-point between Mistral NeMo 12B and Mistral Large 2. This version delivers significant improvements in human alignment, reasoning capabilities, and code over the previous version.

Versatile photorealistic SD 1.5 fine-tune capable of generating a wide range of convincing photographic images.

SDXL fine-tune optimizing for generating photorealistic people, animals, and landscapes. In addition to training, this model is a merge of over 10 other SDXL models aimed at realism.

Update to the Command R series with improved performance for multilingual RAG and tool use. It is better at math, code and reasoning and is competitive with the previous version of the larger Command R+ model.

A popular versatile model finetuned on SDXL by KandooAI and RunDiffusion. This version features improved prompt adherence due to an innovative GPT-4 captioning system built by LEOSAM.

Diffusion transformer model generating 10-second videos at 768x1360 resolution from text prompts using 3D attention mechanisms.

A 3.8 billion parameter multilingual language model with 128K token context window, optimized for reasoning tasks across 22 languages.

A 4.2-billion-parameter multimodal model that processes text and multiple images simultaneously for visual reasoning and document analysis tasks.

An 8-billion parameter language model fine-tuned for multi-turn dialogue, creative writing, and roleplaying using curated conversational datasets and synthetic data.

Merge of multiple SDXL checkpoints and LoRAs, with impressive breadth and consistency in generating photorealistic images.

A 12.2 billion parameter text generation model optimized for creative storytelling, role-playing scenarios, and adventure-based interactive fiction applications.

Flow-based text-to-image model with 6.8 billion parameters featuring enhanced prompt following and multi-aspect ratio support.

A 2-billion parameter diffusion Transformer model that generates up to 6-second videos at 720×480 resolution from text prompts.

Generates textured 3D meshes with material properties from single images in approximately 0.5 seconds using a transformer-based architecture.

FLUX is a 12 billion parameter rectified flow transformer text-to-image model. The "Dev" model is a guidance-distilled version of FLUX.1 [pro] which retains similar quality and prompt adherence capabilities.

FLUX is a 12B parameter rectified flow transformer text-to-image model. The "Schnell" model is the fastest version of FLUX.1 which can generate high-quality images in fewer steps.

A generative video-to-video diffusion model that synthesizes temporally and spatially consistent multi-view video sequences from single input videos.

123 billion parameter model from Paris-based Mistral AI, significantly more capable than its predecessor in code generation, mathematics, reasoning, multilingual support, and function calling.

Open-source text-to-audio synthesis model with 1.21 billion parameters, trained exclusively on Creative Commons data using latent diffusion architecture.

A 12B parameter multi-lingual model that supports function calling built in collaboration with NVIDIA and trained using the new Tekken tokenizer. By some metrics, it is state-of-the-art in its size category. NeMo was trained with quantisation awareness, enabling FP8 inference without any performance loss.

Open-source Mixture-of-Experts model with 236B total parameters specialized for code generation, mathematical reasoning, and programming across 338 languages.

The Llama 3.1 series of open models rivals top closed models in performance. It was trained on over 15 trillion tokens using over 16K H100 GPUs. These models display state-of-the-art capabilities in general knowledge, steerability, math, tool use, and translation.

Llama 3.1 is a new state-of-the-art large language model from Meta.

A 9-billion parameter decoder-only transformer model trained on 8 trillion tokens, designed for text generation and natural language processing tasks.

Open model from Google built from the same research and technology used to create the Gemini models. Well-suited for a variety of text-generation tasks, including question answering, summarization, and reasoning.

Photography-focused SDXL checkpoint with strong prompt adherence and versatile output.

A 16B parameter Mixture-of-Experts model designed for code generation, completion, and reasoning across 338 programming languages with 128K token context length.

A 7.6 billion parameter multilingual decoder-only Transformer model designed for further post-training with 32,000-token context and strong coding capabilities.

A 72-billion parameter transformer model supporting extended context lengths up to 128K tokens and demonstrating strong multilingual capabilities across diverse benchmarks.

A 22.2 billion parameter causal language model trained on over 80 programming languages with support for code generation and fill-in-the-middle completion.

SD 1.5 fine-tune specialized in creating photorealistic portraits of humans.

A 34-billion-parameter decoder-only Transformer model trained on 3.1 trillion bilingual tokens with capabilities in reasoning, mathematics, and code.

A 236-billion parameter mixture-of-experts language model with multi-head latent attention, activating 21 billion parameters per token for bilingual text generation.

A 236-billion parameter Mixture-of-Experts language model that activates only 21 billion parameters per token for efficient multilingual text generation.

A 7-billion parameter language model specialized for code generation, completion, and infilling across multiple programming languages with natural language understanding capabilities.

Mistral instruction-following fine-tune with strong coding abilities, trained on a dataset inspired by the Orca paper and intentionally unaligned / uncensored.

A 3.8 billion parameter instruction-tuned transformer model with 128K token context window, optimized for reasoning, code generation, and structured output.

State-of-the-art 70B foundation model from Meta, trained on over 15 trillion tokens.

Large language model with 8 billion parameters featuring transformer architecture, trained on 15 trillion tokens for text generation and coding tasks.

A sparse mixture-of-experts architecture with 141 billion total parameters, activating 39 billion per token, supporting multilingual tasks and 64,000-token contexts.

A Stable Diffusion 1.5-based checkpoint model designed for photorealistic image generation with simplified prompting and demographic diversity.

A natively multimodal transformer with mixture-of-experts architecture supporting 10 million token context and multilingual text-image understanding capabilities.

Multimodal instruction-tuned model with 17 billion active parameters from 400 billion total, trained on 22 trillion tokens across 200 languages.

103B model from Cohere, with a large 128k context length, trained for enterprise RAG use-cases.

SDXL fine-tune with very strong performance in genearating anime images.

High-ranking 7B LLM trained from Mistral 7B and tuned using RLAIF(Reinforcement Learning from AI Feedback)

Stable Video 3D (SV3D) is a generative model based on Stable Video Diffusion that takes in a still image of an object as a conditioning frame, and generates an orbital video of that object.

35B model from Cohere, with a large 128k context length, trained for enterprise RAG use-cases.

SDXL variant using the Adversarial Diffusion Distillation (ADD) technique to generate images very quickly, with multiple variants for different numbers of steps (1-8) and a more permissive license than SDXL Turbo.

Fine-Tuned on SDXL, this is a general purpose model designed for photos, art, anime, and manga. This Lightning version can generate quality images in few steps.

Foundation open-source 7B LLM from Google, developed using the same technology as their flagship Gemini models.

Based on the SDXL architecture, reportedly produces images that are 3x more aesthetically pleasing by focussing on color, contrast, multi-aspect ratios, and human-centric fine detail.

A vector quantized GAN encoder that compresses 1024×1024 images to 256×256 discrete tokens as part of a three-stage hierarchical text-to-image pipeline.

Intermediate latent super-resolution module that upscales compressed text-conditional representations from Stage C to enable high-fidelity image generation.

Text-to-image diffusion model using three-stage cascaded architecture with 42:1 spatial compression for efficient high-resolution synthesis.

SDXL fine-tune resulting from merging 300+ top community models, strong performance across a wide range of image types.

Video generation model that transforms single images into 25-frame sequences at 1024x576 resolution with controllable motion and camera parameters.

A 24-billion parameter instruction-tuned language model with multilingual capabilities, 32K context window, and optimized low-latency inference performance.

Foundation 72B Qwen 1.5 model from Alibaba Cloud.

Foundation 32B Qwen 1.5 model from Alibaba Cloud.

SDXL fine-tune capable of generating realistic images of people and landscapes.

Mixtral 8X7B fine-tuned on over 1m curated GPT-4 outputs.

SDXL checkpoint optimized for photorealistic generation of a wide range of humans.

A plug-and-play motion module that transforms static Stable Diffusion models into video generators through temporal Transformer architecture.

A fine-tuned Stable Diffusion XL model trained on 2.6 million images for generating cartoon, anime, and anthropomorphic character art.

Mixtral instruction-following fine-tune with strong coding abilities, trained on a dataset inspired by the Orca paper and intentionally unaligned / uncensored.

SDXL model focused on prompt adherence and semantic understanding, with a stated aim of achieving Dalle-3 level understanding of prompts.

Popular SD 1.5 fine-tune with capability to produce detailed images of a versatile breadth of subjects.

Motion module extending Stable Diffusion 1.5 with temporal transformers for text-to-video generation through modular architecture.

Small language model (2.7b) from Microsoft with performance exceeding that of 7b and 13b models on some evaluations.

Foundation 10.7 billion parameter model with performance exceeding that of some 30B models, training using a new methodology called depth up-scaling (DUS).

Spare sparse mixture of experts (MoE) model consisting of 8 "experts" with 7b parameters each, outperforming all Llama 2 models (including 70b) on most benchmarks.

Top fine-tune of MPT-7B, with improved performance over base model in chat, code generation, and summarization.

Seamless is a MMMMT (Massively Multilingual and Multimodal Machine Translation) model that can translate and generate audio in over 100 languages.

A diffusion-based text-to-image model utilizing dual text encoders to generate high-aesthetic 1024x1024 pixel images from natural language prompts.

Mistral 7B fine-tuned on over 1m curated GPT-4 outputs.

SDXL variant using the Adversarial Diffusion Distillation (ADD) technique to generate images very quickly, only requiring 2-6 steps instead of 20-60.

A latent diffusion model that generates short video clips up to 25 frames from single images using temporal convolution and attention layers.

A video generation model that creates coherent sequences from static images or text prompts using latent diffusion architecture.

Small 7B Bi-lingual (English-Mandarin) foundation model.

Large 34B Bi-lingual (English-Mandarin) foundation model with top ranks on many benchmarks.

Widely used transformer model from OpenAI for SST (speech to text), automatic speech recognition (ASR), and speech translation.

MusicGen is a text-to-music autoregressive transformer model capable of generating music samples conditioned on text descriptions or audio prompts.

Magnet is a single-stage, non-autoregressive transformer model for text-to-music and text-to-audio generation, with faster generation that MusicGen.

SD 1.5 ControlNet model for conditioning on an image prompt.

SDXL ControlNet model for conditioning on an image prompt.

Llama 2 7B fine-tuned using the C-RLFT strategy, with performance surpassing GPT 3.5 on some benchmarks.

Mistral 7B fine-tuned on multiple community generated ChatGPT datasets, trained using Direct Preference Optimization (DPO).

Vocos is a fast neural vocoder, Generative Adversarial Network (GAN) objective, designed to synthesize audio waveforms from acoustic features.

Llama 2 7B fine-tune using numerous emerging techniques including SFT, RM, and RLHF, achieving top results on AlpacaEval.

Llama 2 13B fine-tune using numerous emerging techniques including SFT, RM, and RLHF, achieving top results on AlpacaEval.

Popular SD 1.5 photorealism fine-tune with training data weighted on people of asian descent.

Mistral 7B fine-tuned on an open-source dataset inspired by the Microsoft's Orca research paper with performance approaching that of Llama 2 70b.

Base 7.3-billion parameter model from Mistral AI, with superior performance to Llama 2 7b (and 13b, impressively) on nearly all benchmarks.

Llama 2 70B fine-tune using numerous emerging techniques including SFT, RM, and RLHF, achieving top results on AlpacaEval.

Llama 7B fine-tuned for conversation, roleplaying, and storywriting.

Llama 13B fine-tuned for conversation, roleplaying, and storywriting.

Merge of two popular fine-tunes: Pygmalion and Mythomax L2, resulting in a model that is good at roleplaying and instruction-following.

Smaller SDXL ControlNet model for edge detection.

SDXL ControlNet model for depth generation.

Smaller SDXL ControlNet model for depth generation.

SDXL ControlNet model for recoloring images.

SDXL ControlNet model for copying human poses.

Fine-tuned version of Llama 7b, with training data optimized for code generation and completion tasks.

Fine-tuned version of Llama 13b, with training data optimized for code generation and completion tasks.

Fine-tuned version of unreleased Llama 2 34b, with training data optimized for code generation and completion tasks.

Fine-tuned version of Llama 70b, with training data optimized for code generation and completion tasks.

SDXL ControlNet model for edge detection.

An improved variant of MythoMix (MythoLogic-L2 and Huginn merge) using an experimental tensor-type merge technique. Originally a merge of multiple popular Llama 2 fine-tunes, including Hermes, Chronos, and Airoboros. Very popular for rich descriptions, storywriting, and roleplay.

Llama 2 70B fine-tuned on over 300k curated GPT-4 outputs.

Popular SD 1.5 fine-tune with high competence in translating simple text prompts into realistic images of people.

Llama 2 70B model fine-tuned to follow complex instructions using the WizardLM training methodology.

One of the top SD 1.5 variant for generating life-like images of people and objects.

SD 1.5 fine-tune with strong art generation ability and a broad generalist capabilities.

Llama 7B fine-tuned on 125k GPT-4 conversations, with strong performance in chatbot and instruction following use-cases.

A text-to-image diffusion model with 3.5 billion parameters utilizing a two-stage generation pipeline for enhanced image quality and prompt adherence.

Llama 2 13B model fine-tuned to follow complex instructions using the WizardLM training methodology.

SD 1.5 ControlNet model for generating stylized QR codes.

7 billion parameter version of the 2nd iteration in Meta's Llama series of open source models.

13 billion parameter version of the 2nd iteration in Meta's Llama series of open source models.

70 billion parameter version of the 2nd iteration in Meta's Llama series of open source models.

Model resulting for merging 7 different anime-focused SD 1.5 checkpoints.

Photon aims to generate photorealistic and visually appealing images effortlessly.

Llama 13b fine-tuned on over 300k curated GPT-4 outputs, known long high-quality outputs.

Flagship open-source foundation model from Mosaic ML.

Bark is a transformer text-to-audio model for creating multilingual speech and other audio such as music, background noise, and sound effects.

SD 1.5 ControlNet model to replicate the composion of a source image using edge-detection.

SD 1.5 ControlNet trained with pixel-to-pixel instruction.

SD 1.5 ControlNet model to replicate the depth of a source image.

SD 1.5 ControlNet model to detect straight-lines, useful for architecture and man-made objects.

SD 1.5 ControlNet model to replicate the depth of a source image, with additional surface details and geometry.

SD 1.5 ControlNet model for copying human poses.

SD 1.5 ControlNet model for converting sketches to images.

SD 1.5 ControlNet model for detecting and segmenting distinct parts of images to use in the generation.

SD 1.5 ControlNet model to detect soft-edges, especially useful for recoloring and stylizing.

SD 1.5 ControlNet model trained with image inpainting.

SD 1.5 ControlNet model trained with line art generation.

SD 1.5 ControlNet model trained with anime line art generation.

SD 1.5 ControlNet model trained with image shuffling.

SD 1.5 ControlNet model trained with image tiling.

Llama 13B fine-tuned on 125k GPT-4 conversations, with strong performance in chatbot and instruction following use-cases.

SD 1.5 fine-tuned on 124k+ additional images generated with Midjourney v4, leading to results that resemble this other closed-source image generation model.

A 7-billion parameter transformer language model trained on one trillion tokens from publicly available datasets for natural language processing research.

A 13-billion parameter transformer-based language model trained on 1.4 trillion tokens from publicly available datasets, demonstrating competitive performance on various benchmarks.

A 32.5 billion parameter transformer-based language model trained on 1.4 trillion tokens of publicly available data with competitive benchmark performance.

A 65.2-billion parameter transformer language model trained on 1.4 trillion tokens from publicly available datasets for research applications.

AudioLDM is a text-to-audio latent diffusion model inspired by Stable Diffusion, capable of generating sounds effects, human speech, and music.

SD 1.5 fine-tuned on a diverse set of analog images, yielding a vintage photographic look.

A text-to-image diffusion model offering 768×768 resolution generation with depth conditioning, inpainting capabilities, and 4x upscaling functionality.

Demucs is a music source separation model based on a U-Net convolutional architecture, currently capable of separating drums, bass, and vocals from the rest of the accompaniment.

Text-to-image diffusion model trained on LAION dataset subset, generating 512x512 images from natural language prompts using latent space processing.

Tortoise is a text-to-speech model with strong multi-voice capabilities and realistic intonation.

A latent text-to-image diffusion model trained on LAION datasets that generates 512×512 images from natural language prompts using compressed latent space processing.